Quasistatic crack growth in linearized elasticitycvgmt.sns.it/media/doc/paper/3033/quasistatic.pdf · linearized elasticity, however, only an a-priori control on the symmetrized gradient

Quasistatic crack growth in linearized elasticity

Manuel Friedrich1 and Francesco Solombrino2

May 6, 2016

Abstract

In this paper we prove a two-dimensional existence result for a varia-tional model of crack growth for brittle materials in the realm of linearizedelasticity. Starting with a time-discretized version of the evolution drivenby a prescribed boundary load, we derive a time-continuous quasistaticcrack growth in the framework of generalized special functions of boundeddeformation (GSBD). As the time-discretization step tends to 0, the ma-jor difficulty lies in showing the stability of the static equilibrium condition,which is achieved by means of a Jump Transfer Lemma generalizing theresult of [20] to the GSBD setting. Moreover, we present a general com-pactness theorem for this framework and prove existence of the evolutionwithout the necessity of a-priori bounds on the displacements or appliedbody forces.

Keywords. Brittle materials, variational fracture, free discontinuity problems, qua-sistatic evolution, crack propagation.

AMS classification. 74R10, 49J45, 70G75

Contents

1 Introduction 2

2 Preliminaries 72.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.2 Convergence in measure . . . . . . . . . . . . . . . . . . . . . . . 82.3 Function spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.4 Caccioppoli partitions . . . . . . . . . . . . . . . . . . . . . . . . 14

3 The model and statement of the main result 15

1Universitat Wien, Fakultat fur Mathematik, Oskar-Morgenstern-Platz 1, 1090 Wien, Aus-tria. [email protected]

2Zentrum Mathematik, Technische Universitat Munchen, Boltzmannstr. 3, 85747 Garching,Germany. [email protected]

1

4 Mathematical tools 174.1 Piecewise Korn inequality in GSBD . . . . . . . . . . . . . . . . . 174.2 A sharp piecewise Korn inequality in GSBD . . . . . . . . . . . . 204.3 Jump transfer lemma in GSBD . . . . . . . . . . . . . . . . . . . 314.4 A general compactness and existence result . . . . . . . . . . . . . 41

5 Proof of the existence result 46

1 Introduction

The mathematical foundations of the theory of brittle fracture were laid by thework of A. Griffith [28] in the 1920s. The fundamental idea is that the formationof cracks may be seen as the result of the competition between the elastic bulkenergy of the body and the work needed to produce a new crack. This latter ismodelled as a surface energy, which, in its simplest form, is proportional to thesurface measure of the crack via a material constant, called the toughness of thematerial. The rigorous mathematical formulation of the problem, introduced in[25], requires that the function t → (u(t),Γ(t)), associating to each time t a de-formation u(t) of the reference configuration and a crack set Γ(t), is a quasistaticevolution satisfying the following conditions:

• (a) static equilibrium: for every t the pair (u(t),Γ(t)) minimizes the energyat time t among all admissible competitors;

• (b) irreversibility: Γ(s) is contained in Γ(t) for 0 ≤ s < t;

• (c) nondissipativity: the derivative of the internal energy equals the powerof the applied forces.

Remarkable features of this approach are that it is able to show crack initation, aswell as a discontinuous evolution of the crack path, which needs not to be a prioriprescribed. On the other hand, establishing a rigorous mathematical frameworkfor the existence of a continuous-time evolution has proved to be quite a hardtask.

Existence results for continuous-time evolution

The first breakthrough results in this direction are the ones in [16] and [10]tackling in a planar setting the case of anti-plane shear and linearized elasticity,respectively. The evolution is driven by a given prescribed load g(t) on a Dirichletpart ∂DΩ of the boundary of the reference configuration Ω. Namely, in the caseconsidered in [10], the energy associated to a displacement u and a crack Γ isgiven by

E(u,Γ) :=

ˆΩ\Γ

Q(e(u)) dx+H1(Γ) , (1)

2

where Q is a quadratic form acting on the symmetrized gradient e(u). At eachtime t, the deformation u(t), which fulfills the boundary condition u(t) = g(t) on∂DΩ \ Γ(t) has to satisfy the minimality property

E(u(t),Γ(t)) ≤ E(v,Γ) (2)

for all Γ ⊃⋃s<t Γ(s) and all v ∈ LD(Ω \ Γ) with v = g(t) on ∂DΩ \ Γ. Here LD

is the space of displacements with square-integrable strains. The existence of anevolution is proved by following the natural idea, in the context of quasistaticbrittle fracture, of starting with a time-discretized evolution, and then lettingthe time-step go to 0. Namely, for a given time step δ > 0 and n ∈ N, the pair(u(nδ),Γ(nδ)) is inductively defined as a solution for the problem

arg min

ˆΩ\Γ

Q(e(u)) dx+H1(Γ)

(3)

among all cracks Γ ⊃ Γ((n−1)δ) and displacements u ∈ LD(Ω\Γ) with u = g(nδ)on ∂DΩ \ Γ. Notice that the existence for the above minimum problems can beproved under the additional restriction that the admissible cracks have at mosta fixed number of connected components. Indeed, in this case the direct methodproves succesful: crack sets are compact and lower semicontinuous with respectto the Hausdorff topology of sets via Go lab’s Theorem (see [27]), while compact-ness of the displacements is recovered via the Poincare-Korn inequality, uponnoticing that the energy stays invariant under subtraction of rigid movementsin the connected components of Ω \ Γ whose boundary has no intersection with∂DΩ \ Γ.

The aforementioned important restriction plays furthermore a fundamentalrole in overcoming a stability issue, which arises when taking the limit for a timestep δ going to 0. Indeed, if this hypothesis is dropped, the convergence in theHausdorff metric of the approximating cracks Γδ(t) (obtained as piecewise con-stant interpolations of Γ(nδ), n ∈ N) to a set Γ(t) does not imply that piecewiseconstant interpolations of the time-discretized displacements uδ(t) converge to asolution of the minimum problem (3). This issue, which is due to a Neumann-sieve-type phenomenon (see [32]), can be overcome in a planar setting imposingan a-priori bound on the connected components of the cracks and using someresults from the analysis of Neumann problems in varying domains, contained in[8, 12].

To avoid this restriction, a different and more powerful approach has beenproposed in [20], and succesfully applied to the case of anti-plane shear in arbi-trary dimension N . In this case, the reference configuration is an infinite cylinderΩ× R with Ω ⊂ Rn open and bounded, and admissible displacements are of theform (0, . . . , 0, u(x)) where x varies in Ω and the only nonzero component u(x) isscalar-valued. In this case, the linear elastic energy reduces to the Dirichlet en-ergy

´Ω\Γ |∇u|

2 dx and the incremental minimum problems become very similar

3

to the strong formulation of the Mumford-Shah functional in image segmentationproposed in [31]. Inspired by De Giorgi’s weak formulation in the space of specialfunctions of bounded variation SBV (Ω) (see [17, 18]), the authors model cracksets as (union of) jump sets of admissible displacements. The minimum problemsto be solved at every time step essentially reduce (up to some modifications inorder to allow for cracks running alongside the boundary) to

arg min

ˆΩ

|∇u|2 dx+HN−1(Ju \ Γ((n− 1)δ))

, (4)

with Ju denoting the jump set of u, among all displacement satisfying u = g(nδ)on ∂DΩ. Provided one assumes an L∞ bound on the boundary datum, the max-imum principle and Ambrosio’s compactness theorem in SBV (see [1]) ensurewell-posedness for the above problem. If u is a solution thereof, the crack set isthen updated by setting Γ(nδ) := Ju ∪ Γ((n− 1)δ).

A key tool introduced in the paper [20] in order to deal with the above men-tioned stability issues, when the time step δ tends to 0, is the so-called JumpTransfer Lemma. It allows to transer most of the jump set of any function inSBV that lies inside of the jump set of a function u onto that of un, if un isa sequence in SBV weakly converging to u. As a consequence of this lemma,the authors are able to recover a weak form of (2) in the limit. The existenceresult has been later generalized to finite hyperelastic energies and vector-valueddeformations in [15], whereas the existence of a weak quasistatic evolution forthe fully linear elastic model (1) has remained an open issue, due to at least twomajor difficulties.

Challenges for linear elastic models

As a first point, even in the static setting the existence of minimizers for the weakformulation is not clear. A natural attempt of generalising (4) consists indeed inconsidering problems of the type

arg min

ˆΩ

Q(e(u)) dx+HN−1(Ju \ Γ)

, (5)

under some prescribed boundary condition g, in the space SBD of special func-tions of bounded deformation (see [3, 6]), for which a symmetrized gradient andan HN−1-rectifiable jump set are well defined. However, weak sequential com-pactness in SBD requires (see [6, Theorem 1.1]) a uniform bound on the L∞

norm of the sequence, similarly to the SBV -case, which in this setting is notguaranteed along a minimizing sequence, due to the lack of a maximum princi-ple. The addition of lower order terms, related for instance to the action of bulkforces, can at least provide some uniform bound on the Lp norm of the mini-mizing sequences, so that, mimicking a succesful approach to similar problems in

4

spaces of functions of bounded variation, one can recover an existence result inthe space GSBD of generalized special functions of bounded deformation. A cor-rect definition of this space and the investigation of the related compactness andlower semicontinuity properties have proved to be a quite delicate issue, whichhas been overcome only recently in the paper [14]. On the other hand, it wouldbe highly desirable to have an existence result also for the model (5) withoutthe addition of lower order terms. This requires a suitable Korn-type inequalityin GSBD to be available, allowing in some sense to reproduce the steps of theexistence proof for (3) in a weak setting.

The other major issue to be faced in order to give an existence proof of aquasistatic evolution with values in GSBD is the generalisation of the JumpTransfer Lemma to this setting. Actually, the proof strategy devised in [20]cannot be straightforwardly reproduced in this context. Indeed, there the jumpset Ju is written as a countable union of pairwise intersections of level sets ofu. The parts of the corresponding level sets for un lying outside Jun are thenshown to have small length. With this, one can transer onto pieces of thesesets the jump Jφ ∩ Ju for a given competitor φ. In this procedure, the coareaformula and the equiintegrability of ∇un play a crucial role. In the framework oflinearized elasticity, however, only an a-priori control on the symmetrized gradientis available. Again, being able to estimate gradients in terms of their symmetrizedpart via a Korn-type inequality would remove parts of these obstacles and be agood starting point for proving an analog of the lemma in the GSBD setting.

The present paper

This preliminary discussion leads us to the purpose of the present paper. Ourgoal is to provide an existence result, in dimension N = 2, for quasistatic crackgrowth in the sense of Griffith in a linearly elastic material. In Theorem 3.1 weshow the existence of a pair (u(t),Γ(t)), with u(t) ∈ GBSD2(Ω), Ju(t) ⊂ Γ(t),and Γ(t) nondecreasing in time, such that u(t) minimizes

ˆΩ

Q(e(v)) dx+H1(Jv \ Γ(t))

among all v ∈ GSBD2(Ω) satisfying the prescribed time-dependent Dirichletcondition g(t), and the total energy satisfies the energy-dissipation balance

E(u(t),Γ(t)) = E(u(0),Γ(0)) +

ˆ t

0

ˆΩ

Ce(u(s, x)) · e(g(s, x)) ds dx .

In the above equality C is the elastic tensor generating the quadratic form Q,so that the integral term can be interpreted as the virtual work of the appliedboundary load. We also mention that, as it is typical of variational problems in

5

spaces of functions of bounded deformation, the boundary condition has to beunderstood in a relaxed sense (see Section 3 for details).

A starting point for our proof strategy is the use of a piecewise Korn inequalityfor GSBD functions, proved in the planar setting in [23], extending other recentresults in the literature ([21, Theorem 1.1] and [13, Theorem 1.2]). For every1 ≤ p < 2 it allows to control the Lp-norm of a displacement and its gradientin terms of the square norm of the symmetrized gradient, provided a suitablepiecewise infinitesimal rigid motion is subtracted. With this construction thejump set is enlarged, but still controlled by the length of the original jump set.

A major ingredient is then a sharp version of the piecewise Korn inequalityproved in Theorem 4.3. We namely show that the jump set can even only beenlarged by a small length at the prize of having only an L1-control on the gradi-ent. This control, however, involves constants which well behave with respect toscaling and particularly are small on small squares (see Remark 4.4 and Remark4.12).

Equipped with this result, we can prove Lemma 4.13, where, up to an arbi-trarily small error θ, the jump set of a weakly compact sequence (un)n in GSBDis shown to coincide with the one of an L1-compact sequence (vn)n of SBV func-tions, whose limit v contains the jump set of the GSBD limit u again up to asmall error θ. Furthermore, the L1-norm of ∇vn is uniformly small in a tubularneighborhood of the jump set Jv.

This allows to prove a Jump Transfer Lemma also in this setting (Theorem4.9), adapting the arguments of [20, Theorem 2.1]. The reflection procedure thatthe authors use there in order to define the sequence (φn)n corresponding to thecompetitor φ, which is not compatible with a control only on the symmetrizedgradient e(φ), is here replaced by a suitable generalization introduced in [33] andadjusted to our purposes in Lemma 4.10.

The existence proof for the minimum problem (5) requires an additional step,namely a version of the sharp piecewise Korn inequality proved in Theorem 4.3which also takes into account the relaxed boundary conditions. This is provedin Theorem 4.7. With this, we can derive a general compactness result for mini-mizing sequences of the energy (5) drawing some ideas from [21]: while typicallysequences are not compact, it is always possible to pass to modifications by sub-tracting suitable piecewise infinitesimal rigid motions (which do not change theelastic part of the energy) at the expense of arbitrarily small additional fractureenergy. This allows us to construct a minimizing sequence (yn)n which satisfiesthe uniform boundˆ

Ω

ψ(|yn|)dx+

ˆΩ

|e(yn)|2dx+H1(Jyn) ≤M

for an increasing concave continuous functions ψ : [0,∞)→ [0,∞). This bound,in general weaker than any Lp-bound, is enough to apply the compactness resultin [14, Theorem 11.3] deducing the existence of a minimizer (see Theorem 4.15

6

and Theorem 4.16 below). An additional delicate point of the proof is showingthat the function ψ is only depending on the reference configuration Ω and theH1 norm of the boundary displacement g(t), so that, under the usual regularityassumptions on the boundary load, it is independent from the time t along anevolution. This is crucial in the proof of Theorem 5.5 where the global stabilityproperty is derived.

Once this two major hurdles have been fixed, the by now well-known ma-chinery succesfully exploited in [20] and in [15], in the linear antiplane and inthe finite elastic context, respectively, can be adapted to our setting with minormodifications, which we however detail to some extent in Section 5. This leadsto the proof of our main result stated in Theorem 3.1.

As already mentioned, we establish the result only in two dimensions as wemake a heavy use of the piecewise Korn inequality of [23] which has been onlyderived in a planar setting due to technical difficulties, concerning the topologicalstructure of crack geometries in higher dimensions. Additionally, also its gener-alisation to the sharp version (Theorem 4.3) and the case of prescribed boundaryconditions (Theorem 4.7) makes use of estimates holding in a planar setting (seeLemma 2.3, Lemma 4.6, and Lemma 4.8). Without this restrictions, the methodswe use actually hold in any dimension. We therefore believe that our results canbe extended to the N -dimensional case and that the proof provides the principaltechniques being necessary to establish the result in arbitrary space dimension.

2 Preliminaries

We introduce the basic notation and the functional spaces we will use in thepaper in a general N -dimensional setting. The underlying space dimension willbe later specified to be N = 2.

2.1 Notations

The N -dimensional Lebesgue measure and the (N − 1)-dimensional Hausdorffmeasure will be denoted by LN and HN−1, respectively. For the standard normof an Euclidean space we always write | · |. The Euclidean space of quadraticN ×N -matrices and the mutually orthogonal subspaces of symmetric and skew-symmetric matrices will be denoted by RN×N , RN×N

sym , and RN×Nskew , respectively.

Moreover, ξ1 · ξ2 will stand for the scalar product of two vectors ξ1, ξ2 ∈ RN . Thesymmetric difference of two sets is denoted by 4.

For every 1 ≤ p ≤ +∞ the usual notation Lp(E;X) and W 1,p(E;X) will beemployed for Lebesgue and Sobolev spaces of functions from a finite-dimensionalmeasurable set E (also assumed to be open, in the Sobolev case) to a Banachspace X. The norm on Lp(E;X) will be often simply denoted by ‖ · ‖p, wheneverdomain and target space are clear from the context. The brackets 〈·, ·〉 will denote

7

the inner product in L2. For p = 2, we will set as usual H1 in place of W 1,2. Fora Sobolev function g(s) from an interval (a, b) to a Banach space X, the symbolg(s) denotes the a.e. well-defined Frechet derivative of g with respect to s.

For an LN -measurable function f from a measurable set E to an Euclideanspace Ξ and ε > 0, the shortcut |f | ≥ ε is used to denote the superlevel setx ∈ E : |f(x)| ≥ ε. A similar convention is also used for sublevel sets. For abounded, measurable set E ⊂ RN we define

diam(E) = ess sup|x− y| : x, y ∈ E.

The above definition is independent of the particular Lebesgue representative.If U is an open set in RN , and u : U → Rm is a LN -measurable function, u is

said to have an approximate limit a ∈ Rm at a point x ∈ U if and only if

lim%→0+

LN (|u− a| ≥ ε ∩B%(x))

%N= 0 for every ε > 0 ,

where B%(x) is the ball of radius % centered at x. In this case, one writesap limy→x u(y) = a. The approximate jump set Ju is defined as the set of pointsx ∈ U such that there exist a 6= b ∈ Rm and ν ∈ SN−1 := ξ ∈ RN : |ξ| = 1with

ap limy→x

(y−x)·ν>0

u(y) = a , ap limy→x

(y−x)·ν<0

u(y) = b .

The triplet (a, b, ν) is uniquely determined up to a permutation of (a, b) and achange of sign of ν, and is denoted by (u+(x), u−(x), νu(x)). The jump of u isthe function [u] : Ju → Rm defined by [u](x) := u+(x)− u−(x) for every x ∈ Ju.It follows from Lusin’s Theorem that u has u(x) as approximate limit at LN -a.e. x ∈ U , in which case one says that u is approximately continuous at x,and therefore Ju is a LN -null set. Given x ∈ U such that u is approximatelycontinuous at x, an m×N matrix ∇u(x) is said to be an approximate gradientof u at x if and only if

ap limy→x

u(y)− u(x)−∇u(x)(y − x)

|y − x|= 0 .

We say that u has an approximate symmetric differential e(u)(x) ∈ RN×Nsym at x if

ap limy→x

u(y)− u(x)− e(u)(x)(y − x) · (y − x)

|y − x|2= 0 .

2.2 Convergence in measure

Throughout the paper d will denote the distance metrizing the convergence inmeasure for measurable functions from a locally compact N -dimensional measur-able set E to an Euclidean space Ξ, whose explicit expression is given by

d(f, g) :=

ˆE

min|f(x)− g(x)|, 1 dx .

8

It is well-known that a sequence (fn)n satisfies d(fn, f) → 0 if and only if, forevery ε > 0, it holds

LN (E ∩ |fn − f | ≥ ε)→ 0

when n → +∞. We will often make use of the following measure-theoreticalresult from [21]. A short proof is reported for the reader’s convenience.

Lemma 2.1. Let F ⊂ RN with LN(F ) < +∞ and let (sn)n, (tn)n be nonnegative,monotone sequences with sn → ∞ and tn → 0 as n → ∞. Then there is anonnegative, increasing, concave function ψ with

lims→+∞

ψ(s) = +∞ (6)

only depending on F , (sn)n, (tn)n such that for every sequence (un)n ⊂ L1(F ; Ξ)with

‖un‖1 ≤ sn, LN(⋂

m≥n|um − un| ≥ 1

)≤ tn

for all n ∈ N there is a not relabeled subsequence such that

supn≥1

ˆF

ψ(|un|)dx ≤ 1 .

Proof. Let An =⋂m≥n|un − um| ≤ 1 and set B1 = A1 as well as Bn = An \⋃n−1

m=1Bm for all n ∈ N. The sets (Bn)n are pairwise disjoint with∑

n LN(Bn) =

LN(F ). We choose 0 = n1 < n2 < . . . such that∑

1≤n≤niLN (Bn)LN (F )

≥ 1 − 4−i. We

let Bi =⋃ni+1

n=ni+1Bn and observe LN(Bi) ≤ 4−iLN(F ).From now on we consider the subsequence (ni)i∈N and observe that the choice

of (ni)i∈N only depends on the sequence (tn)n. Choose Ei ⊃ Bi such thatLN(Ei) = 4−iLN(F ). Let bi =

sni+1

LN (Ei)+ 2 = 4i

sni+1

LN (F )+ 2 for i ∈ N and note

that (bi)i is increasing with bi → ∞. By an elementary construction (see [21,Lemma 4.1]) we find an increasing concave function ψ : [0,∞) → [0,∞) with

limx→∞ ψ(x) =∞ and ψ(bi) ≤ 2i

LN (F )for all i ∈ N.

For Bi := Ω \⋃nin=1Bn we have LN(Bi) ≤ 4−iLN(F ) and choose Ei ⊃ Bi

with LN(Ei) = 4−iLN(F ). We then obtainsni

LN (Ei)= 4i

sniLN (F )

≤ bi. Now let

l = ni. Using Jensen’s inequality, the definition of the sets Bi, ‖ul‖1 ≤ sl andthe monotonicity of ψ we computeˆ

F

ψ(|ul|) =∑

1≤j≤i−1

ˆBjψ(|ul|)dx+

ˆBiψ(|ul|)dx

=∑

1≤j≤i−1

ˆBjψ(|unj+1

|+ 2)dx+

ˆBiψ(|ul|)dx

≤∑

1≤j≤i−1LN(Ej)ψ

(−ˆEj|unj+1

|+ 2)

+ LN(Ei)ψ(−ˆEi|ul|)

≤∑

1≤j≤i−1LN(F )4−j 2j

LN (F )+ LN(F )4−i 2i

LN (F )≤∑

j∈N2−j = 1.

9

As the estimate is independent of l ∈ (ni)i, this yields´Fψ(|ul|)dx ≤ 1 uniformly

in l, as desired.

Remark 2.2. Let u be a measurable function and (un)n ⊂ L1(F ; Ξ) a sequencesuch that d(un, u)→ 0. Then it follows from the previous lemma that there exista subsequence (unk)k of (un)n and a nonnegative, increasing, concave function ψsatisfying (6), such that

supk≥1

ˆF

ψ(|unk |)dx ≤ 1 .

Indeed, by definition of convergence in measure we can always find a subsequence(unk)k with the property that, setting Ek := |unk − u| ≥ 1

2k , one has LN(Ek) ≤

12k

. Now, for all k ∈ N we have by the triangle inequality that⋃m≥k

|unm − unk | ≥ 1 ⊆⋃m≥k

Em

and therefore

LN(⋃

m≥k|unm − unk | ≥ 1

)≤∑+∞

m=k

1

2m=

1

2k−1.

It then suffices to apply the previous lemma with sk := maxmax1≤i≤k ‖unk‖1, kand tk := 1

2k−1 .

In a 2-dimensional setting, we will often make use of the following simplelemma.

Lemma 2.3. Let A ∈ R2×2skew, b ∈ R2.

(a) There is a universal constant c > 0 independent of A and b such that for

all measurable E ⊂ R2 we have (L2(E))12 |A| ≤ c‖Ax+ b‖L∞(E;R2).

(b) Let F be a bounded measurable subset of R2, δ > 0 and a continuous non-decreasing function ψ : R+ → R+ satisfying (6) be given. Consider a measurablesubset E ⊂ F with L2(E) ≥ δ. Then, if

M ≥ˆE

ψ(|Ax+ b|) dx ,

there exists a constant C only depending on M , δ, ψ and F such that

|A|+ |b| ≤ C . (7)

If ψ(s) = sp for p ∈ [1,∞) we get |A| + |b| ≤ CM1p for a constant C only

depending on δ, p and F .

10

Proof. (a) It suffices to consider the case A 6= 0. If A 6= 0, the assumption

A ∈ R2×2skew implies that A is invertible and that |Ay| =

√2

2|A||y| for all y ∈ R2.

We notice that for all z ∈ R2 there exists x ∈ E with |x − z| ≥ 14diam(E). For

the special choice z = −A−1b we obtain |Ax+ b| = |A (x− z)| =√

22|A||x− z| ≥

√2

8|A|diam(E) which implies the result due to the isodiametric inequality.(b) If A = 0, we have

M

δ≥ ψ(|b|)

and the result follows from (6). If A 6= 0, we set z := −A−1b and λ :=√

δ2π

.

Then we have that L2(E \Bλ(z)) ≥ δ2. Since ψ is nonnegative and increasing we

get

M ≥ˆE

ψ

(√2

2|A||x− z|

)dx

≥ˆE∩B(z,λ)c

ψ

(√2

2|A||x− z|

)dx ≥ δ

2ψ

(√2

2|A|λ

).

It follows by this and (6) that it exists a constant C only depending on M , δ,and ψ such that

|A| ≤ C. (8)

It also follows that |Ax| ≤ C ′ for all x ∈ F , where C ′ is allowed to depend on F ,too. If now |b| ≤ C ′ we are done, otherwise it holds |Ax + b| ≥ |b| − C ′ > 0 forall x ∈ F . The monotonicty of ψ yields then

M

δ≥ ψ(|b| − C ′)

and again (6) implies the conclusion. The case ψ(s) = sp may be proved along

similar lines taking into account that (8) can be replaced by |A| ≤ CM1p for C

independent of M .

2.3 Function spaces

BV- and GSBV-functions. Let an open set U ⊂ RN be given. The spaceBV (U ;RN) consists of the functions u ∈ L1

loc(U ;RN) such that the distributionalgradient Du is a RN×N -valued bounded Radon measure on U . BV -functionshave an approximate differential ∇u(x) at LN -a.e. x ∈ U ([4, Theorem 3.83]),their jump set Ju is HN−1-rectifiable in the sense of [4, Definition 2.57], and νu isa measure-theoretical normal to Ju in the sense of De Giorgi at HN−1-a.e. x ∈ Ju

11

([4, Theorem 3.78]). The space SBV (U ;RN) of special functions of boundedvariation consists of those u ∈ BV (U ;RN) such that

Du = ∇uLN + [u]⊗ νuHN−1bJu .

In the sequel we will make use of the space SBV p(U ;RN), with 1 < p < +∞,defined through:

SBV p(U ;RN) := u ∈ SBV (U ;RN) : ‖∇u‖Lp(U ;RN×N )+HN−1(Ju) < +∞ . (9)

The space GSBV (U ;RN) of the functions of generalized bounded variation isdefined as the set of measurable functions u : U → RN such that ϕ(u) ∈SBVloc(U ;RN) for all ϕ ∈ C1(RN ;RN) with supp(∇φ) ⊂⊂ RN . Also GSBV -functions have an approximate differential ∇u(x) at a.e. x ∈ U and their jumpset Ju is HN−1-rectifiable (see [2]). If 1 < p < +∞, the space GSBV p(U ;RN) isdefined as in (9), with GSBV in place of SBV .

BD-functions. For every u ∈ L1(U ;RN), let Eu be the RN×Nsym -valued distri-

bution on U whose components are defined by Eiju = 12(Djui+Diuj). The space

BD(U) of functions of bounded deformation is the space of all u ∈ L1(U ;RN)such that Eu is a bounded Radon measure on U with values in RN×N

sym . For thegeneral properties of BD(U) we refer to [34] and [3]. In this last paper it is inparticular proved that BD-functions have an approximate symmetric differen-tial e(u)(x) at LN -a.e. x ∈ U , that their jump set Ju is HN−1-rectifiable, andνu is a measure-theoretical normal to Ju in the sense of De Giorgi at HN−1-a.e.x ∈ Ju. The space SBD(U) of special functions of bounded deformation (see[3, 6]) consists of those u ∈ BD(U) such that

Eu = e(u)LN + [u] νuHN−1bJu ,

where denotes the symmetrized tensor product. If 1 < p < +∞, the spaceSBDp(U) is defined as in (9), with SBD in place of SBV , and e(u) in place of∇u.

GBD-functions. We now summarize the definition and some properties ofgeneralized functions of bounded deformation, referring the reader to [14] for moredetails. In the next definition, for fixed ξ ∈ SN−1, we set

Πξ := y ∈ RN : y · ξ = 0 , U ξy := t ∈ R : y + tξ ∈ U for y ∈ Πξ ,

U ξ := y ∈ Πξ : U ξy 6= ∅ .

Definition 2.4. An LN -measurable function u : U → RN belongs to GBD(U) ifthere exists a positive bounded Radon measure λu such that, for all τ ∈ C1(RN)with −1

2≤ τ ≤ 1

2and 0 ≤ τ ′ ≤ 1, and all ξ ∈ SN−1, the distributional derivative

Dξ(τ(u · ξ)) is a bounded Radon measure on U whose total variation satisfies

|Dξ(τ(u · ξ))| (B) ≤ λu(B)

12

for every Borel subset B of U . A function u ∈ GBD(U) belongs to the subsetGSBD(U) of special functions of bounded deformation if in addition for everyξ ∈ SN−1 and HN−1-a.e. y ∈ Πξ, the function uξy(t) := u(y + tξ) belongs toSBVloc(U

ξy ).

By [14, Remark 4.5] one has the inclusionsBD(U) ⊂ GBD(U) and SBD(U) ⊂GSBD(U), which are in general strict. Some relevant properties of functionswith bounded deformation can be generalized to this weak setting: in particu-lar, in [14, Theorem 6.2 and Theorem 9.1] it is shown that the jump set Ju of aGBD-function is HN−1-rectifiable and that GBD-functions have an approximatesymmetric differential e(u)(x) at LN -a.e. x ∈ U , respectively.

Furthermore, the following compactness theorem has been proved, which weslightly adapt for our purposes.

Theorem 2.5. Let Γ be a measurable set with HN−1(Γ) < +∞. Let (yk)k bea sequence in GSBD(U). Suppose that there exist a constant M > 0 and anincreasing continuous functions ψ : [0,∞) → [0,∞) with limx→∞ ψ(x) = +∞such that ˆ

U

ψ(|yk|)dx+

ˆU

|e(yk)|2dx+HN−1(Jyk) ≤M

for every k ∈ N. Then there exist a subsequence, still denoted by (yk)k, and afunction y ∈ GSBD(U) such that

yk → y in measure in U,

e(yk) e(y) weakly in L2(U ;RN×Nsym ),

HN−1(Jy \ Γ) ≤ lim infk→∞

HN−1(Jyk \ Γ).

(10)

Proof. In [14] the assertion has been proved in the case Γ = ∅. We briefly indicatethe necessary adaption for the derivation of (10)(iii) following the argumentationin [15, Theorem 2.8]. If Γ is compact, it suffices to replace Ω by Ω \ Γ. In thegeneral case let K ⊂ Γ compact with H1(Γ \K) ≤ ε. Since Jy \ Γ ⊂ Jy \K andJyk \K ⊂ (Jyk \ Γ) ∪ (Γ \K) we have

H1(Jy \ Γ) ≤ H1(Jy \K) ≤ lim infk→∞H1(Jyk \K)

≤ lim infk→∞H1(Jyk \ Γ) +H1(Γ \K) ≤ lim infk→∞H1(Jyk \ Γ) + ε.

We conclude by letting ε→ 0.

For 1 < p < +∞, the space GSBDp(U), with 1 < p < +∞ is defined through:

GSBDp(U) := u ∈ GSBD(U) : e(u) ∈ Lp(U ;RN×Nsym ) , HN−1(Ju) < +∞ .

(11)We now define a class of displacements with regular jump set. We say that

u ∈ L1(U ;RN) is a displacements with regular jump set if the following properties

13

are satisfied

(i) u ∈ SBV 2(U ;RN),

(ii) Ju =m⋃k=1

Σk, Σk closed connected pieces of C1-hypersurfaces, (12)

(iii) u ∈ H1(U \ Ju;RN).

Displacements with regular jump set are dense in GSBD2(U)∩L2(U ;RN) in thesense given by the following statement, proved in [29] (cf. also [11, Theorem 3,Remark 5.3])).

Theorem 2.6. Let U ⊂ RN open, bounded with Lipschitz boundary. Let u ∈GSBD2(U) ∩ L2(U ;RN). Then there exists a sequence uk of displacements withregular jump set so that

(i) ‖uk − u‖L2(Ω;RN ) → 0

(ii) ‖e(uk)− e(u)‖L2(Ω;RN×Nsym ) → 0,

(iii) HN−1(Juk4Ju)→ 0.

2.4 Caccioppoli partitions

We say that a partition P = (Pj)j of an open set U ⊂ RN is a Caccioppolipartition of U if ∑

jH1(∂∗Pj) < +∞

where ∂∗Pj denotes the essential boundary of Pj (see [4, Definition 3.60]). We saya partition is ordered if LN (Pi) ≥ LN (Pj) for i ≤ j. In the whole article we willalways tacitly assume that partitions are ordered. Moreover, we say that a setof finite perimeter Pj is indecomposable if it cannot be written as P 1 ∪ P 2 withP 1∩P 2 = ∅, LN(P 1),LN(P 2) > 0 andHN−1(∂∗Pj) = HN−1(∂∗P 1)+HN−1(∂∗P 2).The local structure of Caccioppoli partitions can be characterized as follows (see[4, Theorem 4.17]).

Theorem 2.7. Let (Pj)j be a Caccioppoli partition of U . Then⋃j(Pj)

1 ∪⋃

i 6=j(∂∗Pi ∩ ∂∗Pj)

contains HN−1-almost all of U .

Here (P )1 denote the points where P has density one (see again [4, Definition3.60]). Essentially, the theorem states that HN−1-a.e. point of U either belongsto exactly one element of the partition or to the intersection of exactly two sets∂∗Pi, ∂

∗Pj. We now state a compactness result for ordered Caccioppoli partitions(see [4, Theorem 4.19, Remark 4.20]) slightly adapted for our purposes.

14

Theorem 2.8. Let U ⊂ RN bounded, open with Lipschitz boundary. Let Pi =(Pj,i)j, i ∈ N, be a sequence of ordered Caccioppoli partitions of U with

supi≥1

∑j≥1HN−1(∂∗Pj,i) < +∞.

Then there exists a Caccioppoli partition P = (Pj)j and a not relabeled subse-quence such that

∑j≥1 LN (Pj,i4Pj)→ 0 as i→∞.

Proof. In [4] it was proved that Pj,i → Pj in measure for all j ∈ N as i→∞. Webriefly show that this already implies

∑j LN (Pj,i4Pj)→ 0 as i→∞. Let ε > 0

and choose j0 ∈ N sufficiently large such that∑

j<j0LN (Pj) ≥ LN (U)− ε. Then

the convergence in measure implies that for i0 large enough depending on j0 wehave

∑j<j0LN (Pj,i4Pj) ≤ ε for all i ≥ i0. Moreover, this overlapping property

and the choice j0 imply∑

j≥j0 LN (Pj,i) ≤ 2ε for i ≥ i0. Consequently, we find∑

j LN (Pj,i4Pj) ≤ 4ε for i ≥ i0. As ε > 0 was arbitrary, the assertion follows.

3 The model and statement of the main result

In this section we introduce the model we study and we fix the related notations.This preliminary discussion is still conducted in a general N -dimensional setting,while our main result, given at the end of the section, is stated and proved onlyin the planar case N = 2.

We analyze the evolution of a brittle material in the sense of Griffith [28] whosetotal energy consists of a linear elastic bulk term and a surface term proportionalto the (N − 1)-dimensional measure of the crack. The body is under the actionof a time-dependent prescribed boundary displacement g(t) on a relatively openpart ∂DΩ of the boundary (Dirichlet part) of the reference configuration Ω ⊂ RN ,which is supposed to be open, bounded with Lipschitz boundary. The rest of theboundary will be instead assumed to be force-free for simplicity. The variablesof the model are a GSBD-valued displacement u and a (not a priori prescribed)crack Γ with finite HN−1 measure. The uncracked part of the body has a linearelastic stored energy of the form

ˆΩ\Γ

Q(e(u)) dx .

In the above expression e(u) is the approximate symmetrized gradient of u andQ : RN×N

sym → R is the quadratic form associated to a symmetric bounded andpositive definite stiffness tensor C : RN×N

sym → RN×Nsym , that is

Q(e) :=1

2Ce : e , (13)

15

with the colon denoting the Euclidean product between matrices.The prescribed boundary displacement g is a time dependent function

g ∈ W 1,1loc ([0,+∞);H1(RN ;RN)). As it is typical for the weak formulation of evo-

lutionary problems in spaces of functions of bounded deformation, the boundarycondition will be relaxed as follows. We will assume that it exists an open,bounded Lipschitz set Ω′ ⊃ Ω such that

Ω′ ∩ Ω = ∂DΩ Ω′ \ Ω has Lipschitz boundary (14)

and impose, for every time t, that an admissible displacement u(t) satisfies u(t) =g(t) a.e. in Ω′ \ Ω. A competing crack may choose indeed to run alongside ∂DΩ,in which case the boundary condition is not attained in the sense of traces, atthe expense of a crack energy.

The energy of a crack Γ ⊂ Ω will be proportional to its (N − 1)-dimensionalHausdorff measure, namely of the form

κHN−1(Γ ∩ Ω′) ,

where the material parameter κ represents the toughness of the material. Withinthis choice, and because of (14), formation of cracks along ∂DΩ is penalized,while no energy is spent for a crack sitting on the load-free part of the boundary∂Ω \ ∂DΩ. In the following we will set κ = 1 without loss of generality.

The quasistatic evolution problem associated to the model with the prescribedboundary displacement g(t) consists in finding a displacement and crack path(u(t),Γ(t)) with Ju(t) ⊂ Γ(t) ⊂ Ω and u(t) = g(t) a.e. in Ω′ \ Ω such that Γ(t) isirreversible, namely Γ(t) ⊃ Γ(s) whenever t > s, and the following two conditionshold:

• global stability. For each t, u(t) minimizesˆ

Ω

Q(e(v)) dx+HN−1(Jv \ Γ(t)) (15)

among all v ∈ GSBD2(Ω′) such that v = g(t) on Ω′ \ Ω;

• energy-dissipation balance. The total energy

E(t) :=

ˆΩ

Q(e(u(t))) dx+HN−1(Γ(t) ∩ Ω′) (16)

is absolutely continuous and satisfies

E(t) = E(0) +

ˆ t

0

〈σ(s), e(g(s))〉 ds (17)

for all t > 0, where σ(s) = Ce(u(s)), and 〈·, ·〉 is the duality pairing inL2(Ω;RN×N

sym ).

16

Notice that even for a given Γ, the existence of a minimizer for the problemconsidered in (15) is a nontrivial issue, which we are able to overcome for themoment only in the planar case N = 2 (Theorem 4.16). Indeed, in the planarcase we are able to show the existence of a quasistatic evolution according to thefollowing statement, which constitutes the main result of the paper.

Theorem 3.1. Let N = 2. Let Ω ⊂ Ω′ be bounded domains in R2 with Lipschitzboundary satisfying (14), g ∈ W 1,1

loc ([0,+∞);H1(R2;R2)), and consider Q as in(13). Then, for all t ≥ 0 it exists an H1-rectifiable crack Γ(t) ⊂ Ω and a fieldu(t) ∈ GSBD2(Ω′) such that

• Γ(t) is nondecreasing in t;

• u(0) minimizes ˆΩ

Q(e(v)) dx+H1(Jv)

among all v ∈ GSBD2(Ω′) such that v = g(0) on Ω′ \ Ω;

• for all t > 0, u(t) satisfies the global stability (15) for N = 2;

• Ju(0) = Γ(0) and Ju(t) ⊂ Γ(t) up to a set of H1-measure 0.

Furthermore, the total energy E(t) defined by (16) satisfies the energy dis-sipation balance (17). Finally, for any countable, dense subset I ⊂ [0,+∞)containing zero, we have

Γ(t) =⋃

τ∈I , τ≤t

Ju(τ)

for all t > 0.

4 Mathematical tools

In this section we discuss the mathematical tools, that we need in order to proveTheorem 3.1. Namely, we prove two major results, the GSBD version of theJump Transfer Lemma (Theorem 4.9) and the existence of minimizers for theincremental problems (Theorem 4.16). Here and henceforth we will call infinites-imal rigid motion an affine mapping of the form aA,b := Ax + b with A ∈ R2×2

skew

and b ∈ R2.

4.1 Piecewise Korn inequality in GSBD

In this section we recall and comment a piecewise Korn inequality for GSBDfunctions, proved in the planar setting in [23] and being one of the major ingre-dients of our proofs. It implies in particular a density result Theorem 4.2 whichin the planar case improves upon Theorem 2.6.

17

Theorem 4.1. Let Ω ⊂ R2 open, bounded with Lipschitz boundary. Let p ∈[1, 2). Then there is a constant c = c(p) > 0 and Ckorn = Ckorn(p,Ω) > 0 suchthat for each u ∈ GSBD2(Ω) there is a Caccioppoli partition Ω =

⋃∞j=1 Pj and

corresponding infinitesimal rigid motions (aj)j = (aAj ,bj)j such that

v := u−∑∞

j=1ajχPj ∈ SBV p(Ω;R2) ∩ L∞(Ω;R2)

and

(i)∑∞

j=1H1(∂∗Pj) ≤ c(H1(Ju) +H1(∂Ω)),

(ii) ‖v‖L∞(Ω;R2) ≤ Ckorn‖e(u)‖L2(Ω;R2×2sym),

(iii) ‖∇v‖Lp(Ω;R2×2) ≤ Ckorn‖e(u)‖L2(Ω;R2×2sym).

(18)

Below in Section 4.2 we prove a refined version of Theorem 4.1 which (a)provides a sharp estimate for the boundary of the partition in (18)(i) and (b)takes into account boundary data. This refined result will then be fundamentalin proving the jump transfer lemma and the existence theorem for the time-incremental minimum problems.

Proof. For a complete proof see [23]. We only add some short comments for thereader’s convenience. The result is proved at first for displacements with regularjump set and up to a small exceptional set.

The general strategy is to identify the regions of various mesoscopic sizeswhere Ju is too large and to apply a Korn inequality for functions with smalljump set (see [22]) on the complement of these sets. This allows to construct apartition of Ω into simply connected sets such that on each component of thepartition the configuration can be modified to a Sobolev function whose distancefrom u can be controlled outside a small exceptional set.

Then using the main result of [24] one passes to another refined partitionconsisting of John domains with uniform John-constant on which the classicalKorn inequality in W 1,p can be used to derive (18)(iii). To deduce (18)(ii) from(18)(iii), the domain is partitioned into level sets by an application of the coareaformula in BV (see also the proof of Proposition 6.2 in [7]): after possibly sub-tracting another piecewise constant function, a bound on the L∞-norm of v canbe obtained with a constant Ckorn which does not depend on the L∞-norm ofthe displacement u with regular jump set, thus permitting later an extension bydensity.

To establish the result for displacements with regular jump set the aboveprocedure is repeated on various scales of mesoscopic size getting progressivelysmaller. After that, the inequality is extended to GSBD2(Ω). For functions ad-ditional lying in L2(Ω) here and in the following one may apply the density resultTheorem 2.6. In the general case u ∈ GSBD2(Ω) we use a variant of Cham-bolle’s density result stated in [23, Corollary 5.5]: for a given u ∈ GSBD2(Ω)

18

there exists a sequence (uk)k of displacements with regular jump set such thatfor a universal constant c > 1 one has d(uk, u)→ 0 and

‖e(uk)‖L2(Ω) ≤ c‖e(u)‖L2(Ω), H1(Juk) ≤ cH1(Ju) + cH1(∂Ω). (19)

Applying Theorem 4.1 on each uk, we find a sequence of (ordered) Caccioppolipartitions Ω =

⋃∞j=1 P

kj and infinitesimal rigid motions (akj )j = (aAkj ,bkj )j such that

vk := uk −∑∞

j=1akjχPkj ∈ SBV

p(Ω;R2) ∩ L∞(Ω;R2)

satisfies (18). We then have∑∞

j=1H1(∂∗P k

j ) ≤ c(H1(Juk) +H1(∂Ω)) ≤ c(H1(Ju) +H1(∂Ω)) .

Therefore, Theorem 2.8 implies the existence of a partition Ω =⋃j Pj and a (not

relabeled) subsequence such that χPkj → χPj in L1(Ω), when k →∞, for all j ∈ Nand such that, passing to the limit in (18)(i) via the lower semicontinuity of theperimeter and (19), we get∑∞

j=1H1(∂∗Pj) ≤ c(H1(Ju) +H1(∂Ω)).

To the sequence vk we can apply Ambrosio’s compactness Theorem [4, Theorem4.36] recovering the existence of a weak limit v ∈ SBV p(Ω;R2) ∩ L∞(Ω;R2) ofvk for which, again by (19)

‖v‖L∞(Ω;R2) ≤ C‖e(u)‖L2(Ω;R2×2sym), ‖∇v‖Lp(Ω;R2×2) ≤ C‖e(u)‖L2(Ω;R2×2

sym).

The only thing to be shown is therefore the existence of rigid motions (aj)j =(aAj ,bj)j such that

u− v =∑∞

j=1ajχPj .

Since Pj is a partition of Ω, this is equivalent to

(u− v)χPj = ajχPj (20)

a.e. in Pj, for all j ∈ N. Clearly, if L2(Pj) = 0 it suffices to set aj = 0. If insteadL2(Pj) > 0, then it exists δ > 0 independently of k such that L2(P k

j ) ≥ δ. Sincethe sequence

akjχPkj = (uk − vk)χPkjis converging in measure to (u− v)χPj , by Remark 2.2 we can assume that thereexists a positive nondecreasing continuous function ψ satisfying (6) such that´Pkjψ(|akj |) dx ≤ 1. By Lemma 2.3 we infer that akj are bounded in W 1,∞(Ω;R2)

for a constant depending on δ, and thus on j, and Ω, but not on k. If aj isthen a uniform limit point of akj , we have akjχPkj → ajχPj in L1(Ω;R2). By the

convergence of akjχPkj to (u−v)χPj , this implies (20) and concludes the proof.

19

Exploiting the above inequality, the following density result for GSBD func-tions has been proved in [23].

Theorem 4.2. Let Ω ⊂ R2 open, bounded with Lipschitz boundary. Let u ∈GSBD2(Ω). Then there exists a sequence uk of displacements with regular jumpset such that

(i) d(uk, u)→ 0

(ii) ‖e(uk)− e(u)‖L2(Ω;R2×2sym) → 0,

(iii) H1(Juk4Ju)→ 0.

Note that in contrast to the original density result reported in Theorem 2.6the assumption that u ∈ L2(Ω) is not needed in the planar setting.

4.2 A sharp piecewise Korn inequality in GSBD

In this section we derive a piecewise Korn inequality with a sharp estimate forthe surface energy and also prove a version taking Dirichlet boundary conditionsinto account.

Theorem 4.3. Let Ω ⊂ R2 open, bounded with Lipschitz boundary and 0 < θ < 1.Then there is a universal constant c > 0, some CΩ = CΩ(Ω) > 0 and someCθ,Ω = Cθ,Ω(θ,Ω) > 0 such that the following holds: For each u ∈ GSBD2(Ω) wefind uθ ∈ SBV (Ω;R2)∩L∞(Ω;R2) such that u 6= uθ is a set of finite perimeterwith

(i) L2(u 6= uθ) ≤ cθ(H1(Ju) +H1(∂Ω))2,

(ii) H1((∂∗u 6= uθ ∩ Ω) \ Ju) ≤ cθ(H1(Ju) +H1(∂Ω)),(21)

a (finite) Caccioppoli partition Ω =⋃Ii=0 Pi, and corresponding infinitesimal rigid

motions (ai)Ii=0 such that v := uθ −

∑Ii=0 aiχPi ∈ SBV (Ω;R2) ∩ L∞(Ω;R2) is

constant on P0 and

(i)∑I

i=0H1((∂∗Pi ∩ Ω) \ Ju) ≤ cθ(H1(Ju) +H1(∂Ω)),

(ii) L2(Pi) ≥ CΩθ2 for all 1 ≤ i ≤ I, L2(u 6= uθ4P0) = 0,

(iii) ‖v‖L∞(Ω;R2) + ‖∇v‖L1(Ω;R2×2) ≤ Cθ,Ω‖e(u)‖L2(Ω;R2×2sym).

(22)

Note that the refined estimate (22)(i) comes at the expense of the fact thatwe have to pass to a slightly modified function (see (21)) and that in (21)(iii)only the L1-norm of the derivative is controlled.

Remark 4.4. Let CQ1 and Cθ,Q1 be the constants in Theorem 4.3 for the unitsquare Ω = Q1 = (0, 1)2. Using a rescaling argument, (22)(ii),(iii) in Theorem4.3 applied for any square Ω = Q ⊂ R2 read as

(i) L2(Pi) ≥ CQ1L2(Q)θ2 for 1 ≤ i ≤ I,

(ii) ‖v‖L∞(Q;R2) + (diam(Q))−1‖∇v‖L1(Q;R2×2) ≤ Cθ,Q1‖e(u)‖L2(Q;R2×2sym).

(23)

20

Below after the proof of Theorem 4.3 we briefly indicate how Remark 4.4can be derived from (22) for convenience of the reader. As a preparation weformulate two lemmas. Recall the notion of decomposable sets in Section 2.4 andthe definition of diam in Section 2.1.

Lemma 4.5. Let B ⊂ R2 an indecomposable, bounded set with finite perimeter.Then diam(B) ≤ H1(∂∗B).

The proof can be found in [30, Propostion 12.19, Remark 12.28]. The followinglemma investigates some properties of the jump set of a piecewise-defined functionon the interface of two sets of finite perimeter.

Lemma 4.6. Let Ω ⊂ R2 open, bounded and y ∈ SBV (Ω;R2) ∩ L∞(Ω;R2). LetP1, P2 ⊂ Ω be sets of finite perimeter and ai = aAi,bi, i = 1, 2, infinitesimal rigidmotions. Then there is a ball B ⊂ R2 with

(i) diam(B) ≤ 4diam(P2) ‖a1 − a2‖−1L∞(P2;R2)

∑i=1,2‖y − ai‖L∞(Pi;R2),

(ii) H1((∂∗P1 ∩ ∂∗P2) \ (B ∪ Jy)

)= 0.

Proof. We define γ = ‖a1 − a2‖L∞(P2;R2) and δ =∑

i=1,2 ‖y − ai‖L∞(Pi;R2) for

shorthand. First, if δ ≥ 12γ, we can choose B as a ball containing P2 with

diam(B) ≤ 2diam(P2). Consequently, it suffices to assume δ < 12γ.

For i = 1, 2 we denote by Tiy the trace of y on ∂∗Pi, which exists by [4,Theorem 3.77] and satisfies

|Tiy(x)− ai(x)| ≤ ‖y − ai‖L∞(Pi;R2) for H1-a.e. x ∈ ∂∗Pi.

Assume the statement was wrong. Then we would find two points x1, x2 with|x1−x2| > 4γ−1δdiam(P2) such that x1, x2 ∈ (∂∗P1∩∂∗P2)\Jy and for i, j = 1, 2

|Tiy(xj)− ai(xj)| ≤ ‖y − ai‖L∞(Pi;R2).

Since x1, x2 /∈ Jy and thus T1y(x1) = T2y(x1), T1y(x2) = T2y(x2) we compute

|a1(xj)− a2(xj)| ≤ |T1y(xj)− a1(xj)|+ |T2y(xj)− a2(xj)| ≤ δ

for j = 1, 2. Combining the estimates for j = 1, 2 we get

|x1 − x2||A1 − A2| ≤ 2|(A1 x1 + b1)− (A2 x1 + b2)− (A1 x2 + b1) + (A2 x2 + b2)|≤ 2(|a1(x1)− a2(x1)|+ |a1(x2)− a2(x2)|) ≤ 2δ

and therefore |A1 − A2| ≤ 12(diam(P2))−1γ as well as

γ = ‖a1 − a2‖L∞(P2;R2) ≤ |a1(x1)− a2(x1)|+ diam(P2)|A1 − A2| ≤ δ + 12γ,

which contradicts γ > 2δ.

21

Proof of Theorem 4.3. Let u ∈ GSBD(Ω) be given and set for shorthand E =‖e(u)‖L2(Ω;R2×2

sym) and J ′u = Ju ∪ ∂Ω. Without restriction we can assume θ−1 ∈ Nand that Ω is connected as otherwise the following arguments are applied foreach connected component of Ω. Moreover, we may suppose that H1(Ju) ≤(θ−1L2(Ω))

12 as otherwise the assertion trivially holds with uθ = 0.

In the following c > 0 stands for a universal constant and CΩ = CΩ(Ω) > 0,Cθ,Ω = Cθ,Ω(θ,Ω) > 0 represent generic constants which may vary from lineto line. We may further assume that θ is chosen (depending on Ω) such thatθ ≤ 1

16C−1

korn, where Ckorn is the constant from (18).

(a) (b)

(c) (d)

P ′1

P ′2

P ′3P ′4

P ′5

P ′6

7 89

10

11

P 11

P 12

P 13

P 14

R1

P 21

P 22

P 23

P 24

P1

P2

P1

Figure 1: Illustration of the constructions in the proof of Theorem 4.3. (a) Thepartition (P ′j)

11j=1 is sketched (for convenience for small components only the indices

are given). Note that in general the jump set (depicted in light gray) is not a subsetof⋃11j=1 ∂

∗P ′j . (b) The large components of (P 1j )6j=1 are given by P 1

1 = P ′1 ∪ P ′9, P 12 =

P ′2∪P ′10, P 13 = P ′3∪P ′8, P 1

4 = P ′4 (i.e. I ′ = 4), the exceptional set is R1 = P ′6∪P ′11 and thesmall components are P ′5, P ′7. Observe that P 1

1 , depicted in light gray, is not connected.(c) The union of balls R2 is illustrated and the set Ωgood =

⋃4j=1 P

1j \R2 =

⋃4j=1 P

2j is

given in light gray. (d) In this example we have R3 = ∅. The set Ωbad is depicted indark gray and Ω \ Ωbad = P 3

1 ∪ P 32 = P1 ∪ P2 consists of two components, i.e. I ′′ = 2.

We further have Z1 = ∅, Z2 = (1, 2), (1, 3), (1, 4), (3, 4) and Z3 = (2, 3), (2, 4).

Step 0 (Overview of the proof). The general idea behind the proof is to suitablymodify the infinitesimal rigid motions provided by Theorem 4.1 so that all thesets Pj of the Caccioppoli partition are almost completely disconnected by Ju: bythis we mean that the interface between different components will be containedin the jump set of u up to a small (in area and perimeter) exceptional set. Indoing this, we must anyway be able not to lose the estimate in (21)(iii). Theseare the main observations that allow us to pursue this strategy:

22

(O1) If the L∞ distance between two infinitesimal rigid motions aj1 and aj2 , thatare subtracted from u on two sets Pj1 and Pj2 , respectively, lies below afixed threshold depending on the error parameter θ (see (25)(iii)), we canreplace aj2 with aj1 on Pj2 . Indeed, by construction and using Lemma2.3(a), (21)(iii) will still hold up to suitably enlarging Cθ,Ω.

(O2) If the L∞ distance between two infinitesimal rigid motions aj1 and aj2 , thatare subtracted from u on two sets Pj1 and Pj2 , respectively, lies above an(even larger) fixed threshold depending on θ (see (25)(iv)), using Lemma4.6 the interface between Pj1 and Pj2 not contained in Ju can be coveredby a small ball. This will lead to neglecting a small exceptional set withsmall perimeter, provided this is not done ‘too often’. Some combinatorialarguments will indeed be needed (cf., for instance, the derivation of (31)later in the proof).

(O3) On neighboring components Pj1 and Pj2 , whose size lies above a fixedthreshold depending on θ, and that are not almost completely disconnectedby Ju, the L∞ estimate in (18)(iii), the continuity of u on part of the in-terface, together with Lemma 2.3(a), allow us to estimate the L∞ distancebetween the corresponding infinitesimal rigid motions aj1 and aj2 basicallyonly in terms of θ, and therefore we may apply (O1) to remove the artifi-cially introduced boundaries.

Guided by these observations, the proof is organized as follows. In Step Iwe reorganize the partition given by Theorem 4.1 into large sets, of size at leastθ2L2(Ω), small sets, covering only a small part of Ω and a rest set, denoted byR1, which has small perimeter (see (24)). Using (O1) the partition has now theproperty that the infinitesimal rigid motions given on large and small components,respectively, differ very much (see (25)(iv)). This is the starting point for Step II,where, in the spirit of (O2), we show that the part of the interfaces between largeand small components not contained in Ju can be covered by an exceptionalset which is small in area and perimeter. In Step III we then investigate thedifference of the infinitesimal rigid motions given on large components, againemploying Lemma 4.6 to completely disconnect various components, and using(O3) on the others. In Step IV we collect all estimates and conclude the proof.

Step I (Identification of large components). The goal of this step is to define aset R1 ⊂ Ω with

H1(∂∗R1) ≤ θH1(J ′u), L2(R1) ≤ cθ2(H1(J ′u))2, (24)

an (ordered) Caccioppoli partition Ω\R1 =⋃∞j=1 P

1j and corresponding infinites-

imal rigid motions (a1j)j such that v1 := u −

∑j≥1 a

1jχP 1

jsatisfies for an index

23

I ′ ∈ N with I ′ ≤ θ−2 and some Kθ ∈ N, Kθ ≤ θ−1,

(i) L2(P 1j ) ≥ θ2L2(Ω) for all 1 ≤ j ≤ I ′, L2

(Ω \

⋃I′

j=1P 1j

)≤ cθ(H1(J ′u))

2,

(ii)∑

j≥1H1(∂∗P 1

j ) ≤ cH1(J ′u),

(iii) ‖v1‖L∞(Ω\R1;R2) ≤ 2Ckornθ−4KθE , ‖∇v1‖L1(Ω\R1;R2×2) ≤ Cθ,ΩE ,

(iv) min1≤i≤I′ ‖a1i − a1

j‖L∞(P 1j ;R2) ≥ θ−4(Kθ+1)E for all j > I ′. (25)

Moreover, the sets (P 1j )j>I′ are indecomposable, while the sets (P 1

j )I′j=1 are pos-

sibly not indecomposable.We first apply Theorem 4.1 to find an ordered Caccioppoli partition (P ′j)j≥1

of Ω and corresponding infinitesimal rigid motions (a′j)j = (aA′j ,b′j)j such that

v′ := u−∑

j≥1 a′jχP ′j ∈ SBV (Ω;R2) ∩ L∞(Ω;R2) satisfies (18), in particular

‖v′‖L∞(Ω;R2) + ‖∇v′‖L1(Ω;R2×2) ≤ CkornE . (26)

Without restriction we assume that the sets (P ′j)j≥1 are indecomposable. LetI ′ ∈ N be the largest index such that L2(P ′I′) ≥ θ2L2(Ω). (Recall that thepartition is assumed to be ordered.) Then I ′ ≤ θ−2 and by the isoperimetricinequality and (18)(i)

(i)∑

j≥1(L2(P ′j))

12 ≤ c

∑j≥1H1(∂∗P ′j) ≤ cH1(J ′u) ≤ Cθ,Ω,

(ii)∑

j>I′L2(P ′j) ≤ θ(L2(Ω))

12

∑j>I′

(L2(P ′j))12

≤ cθ(H1(∂Ω))12

∑j>I′H1(∂∗P ′j) ≤ cθ(H1(J ′u))

2,

(27)

where in the last step of (i) we used the assumption H1(Ju) ≤ (θ−1L2(Ω))12 . We

introduce a decomposition for the small components according to the differenceof infinitesimal rigid motions as follows. For k ∈ N we introduce the set of indices

J 0 = j > I ′ : min1≤i≤I′ ‖a′j − a′i‖L∞(P ′j ;R2) ≤ Eθ−4,

J k = j > I ′ : Eθ−4k < min1≤i≤I′ ‖a′j − a′i‖L∞(P ′j ;R2) ≤ Eθ−4(k+1)(28)

and define sk =∑

j∈J k H1(∂∗P ′j) for k ∈ N0. In view of (27)(i) we find some

Kθ ∈ N, Kθ ≤ θ−1, such that sKθ ≤ cθH1(J ′u).We let R1 :=

⋃j∈JKθ P

′j and the choice of Kθ together with the isoperimetric

inequality shows (24). We introduce the Caccioppoli partition (P 1j )j≥1 of Ω \R1

by combining different components of (P ′j)j≥1. We now decompose the indices in⋃Kθ−1k=0 J k into sets J ′i with

⋃I′

i=1 J ′i =⋃Kθ−1k=0 J k according to the following rule:

an index j ∈ J k is assigned to J ′i whenever i is the smallest index such that theminimum in (28) is attained.

24

Define the large components P 1i = P ′i ∪

⋃j∈J ′i

P ′j for 1 ≤ i ≤ I ′ and by (P 1i )i>I′

we denote the small componentsP ′j : j > I ′, j ∈

⋃∞

k=Kθ+1J k. (29)

Then (25)(i) holds by (27)(ii) and we see that the sets (P 1j )j>I′ are indecompos-

able. Likewise, (25)(ii) follows from (27)(i). Moreover, we define a1j = a′j for

1 ≤ j ≤ I ′ and let a1j = a′kj for j > I ′, where kj ∈ N such that P 1

j = P ′kj . We

introduce v1 = u−∑

j≥1 a1jχP 1

jand observe that by (26), (28) and the definition

of P 1j for 1 ≤ j ≤ I ′ we have

‖v1‖L∞(P 1j ;R2) ≤ ‖v′‖L∞(P 1

j ;R2) + ‖v1 − v′‖L∞(P 1j ;R2) ≤ CkornE + θ−4KθE

≤ 2Ckornθ−4KθE .

Moreover, by Lemma 2.3, (26), (27)(i), (28) and the definition of J ′i we find∑I′

i=1‖∇v1‖L1(P 1

i ;R2×2) ≤∑I′

i=1

(‖∇v′‖L1(P 1

i ;R2×2) +∑

j∈J ′iL2(P ′j)|A′j − A1

i |)

≤ ‖∇v′‖L1(Ω;R2×2) +I′∑i=1

∑j∈J ′i

(L2(P ′j))12‖a′j − a′i‖L∞(P ′j ;R2)

≤ CkornE + θ−4KθE∑

j≥1(L2(P ′j))

12 ≤ Cθ,ΩE .

Note that the last constant Cθ,Ω indeed only depends on θ and Ω since Kθ ≤ θ−1

and Ckorn only depends on Ω. The last two estimates together with (26) show(25)(iii). Finally, the definition of the small components in (29) together with(28) implies (25)(iv).

Step II (Interface between large and small components). We now show that there

is a union of balls R2 ⊂ Ω and a Caccioppoli partition⋃I′

j=1 P2j of Ωgood :=⋃I′

j=1(P 1j \ R2) and corresponding infinitesimal rigid motions (a2

j)I′j=1 such that

with v2 := u−∑I′

j=1 a2jχP 2

jwe have

(i) L2(Ω \ Ωgood) ≤ cθ(H1(J ′u))2,

(ii) H1(∂∗Ωgood \ J ′u) ≤ cθH1(J ′u),

(iii)∑I′

j=1H1(∂∗P 2

j ) ≤ cH1(J ′u),

(iv) ‖v2‖L∞(Ωgood;R2) + ‖∇v2‖L1(Ωgood;R2×2) ≤ Cθ,ΩE .

(30)

First, for each 1 ≤ i ≤ I ′ and j > I ′ we apply Lemma 4.6 for P1 = P 1i and

P2 = P 1j and obtain a ball Bi,j with H1((∂∗P 1

i ∩∂∗P 1j )\ (Bi,j ∪Ju)) = 0 such that

by (25)(iii),(iv)

diam(Bi,j) ≤ 16Ckorn diam(P 1j ) · θ−4Kθ · (θ−4(Kθ+1))−1 ≤ θ3 diam(P 1

j ),

25

where the last step follows from the fact that θ ≤ 116C−1

korn. Then by Lemma 4.5and the fact that P 1

j is indecomposable (see below (25)) we get diam(Bi,j) ≤θ3H1(∂∗P 1

j ).Define R2 =

⋃i≤I′<j Bi,j and compute by (25)(ii) and I ′ ≤ θ−2 (cf. (25)(i))∑

i≤I′<jH1(∂Bi,j) ≤ θ3I ′

∑j>I′H1(∂∗P 1

j ) ≤ cθH1(J ′u). (31)

Then the isoperimetric inequality yields L2(R2) ≤ cθ2(H1(J ′u))2 and this together

with (25)(i) shows (30)(i). Let P 2j = P 1

j \ R2 and a2j = a1

j for 1 ≤ j ≤ I ′. Then(30)(iii) follows from (25)(ii) and (31). To see (30)(ii), we calculate by Theorem2.7, (24) and (31) recalling that Ωgood ∪

⋃j>I′(P

1j \ R2) ∪ (R1 \ R2) ∪ R2 is a

partition of Ω

H1(∂∗Ωgood \ (Ju ∪ ∂Ω)) ≤∑i≤I′<j

(H1((∂∗P 1

i ∩ ∂∗P 1j ) \ (Ju ∪Bi,j)

)+H1(∂Bi,j)

)+H1(∂∗R1) ≤ 0 + cθH1(J ′u) = cθH1(J ′u).

Finally, (30)(iv) follows from (25)(iii), the definition of v2 and the fact thatKθ ≤ θ−1.

Step III (Interface between large components). We now investigate the differenceof the infinitesimal rigid motions (a2

j)I′j=1. We show that there is a union of

balls R3 ⊂ Ω and a Caccioppoli partition Ωgood \ R3 =⋃I′′

i=1 P3i with I ′′ ≤ I ′

and corresponding infinitesimal rigid motions (a3i )I′′i=1 such that with v3 := u −∑I′′

i=1 a3iχP 3

iwe have

(i) H1(∂∗R3) ≤ cθH1(J ′u), L2(R3) ≤ cθ2(H1(J ′u))2,

(ii)∑I′′

i=1H1(∂∗P 3

i \ J ′u) ≤ cθH1(J ′u),

(iii) ‖v3‖L∞(Ωgood\R3;R2) + ‖∇v3‖L1(Ωgood\R3;R2×2) ≤ Cθ,ΩE .

(32)

In the following we denote the constant given in (30)(iv) by C = C(θ,Ω) todistinguish it from other generic constants Cθ,Ω. We introduce the set of indicesZ1 = 1 ≤ j ≤ I ′ : diam(P 2

j ) ≤ θ3H1(∂Ω) 3 and let Z2 = (i, j) : 1 ≤ i < j ≤I ′, i, j /∈ Z1 be the collection of pairs with

maxk=i,j ‖a2i − a2

j‖L∞(P 2k ;R2) > Cθ−5E . (33)

Finally, let Z3 = (i, j) : 1 ≤ i < j ≤ I ′, i, j /∈ Z1, (i, j) /∈ Z2.3The introduction of Z1 is only a technical point due to the fact that by the previous step

some large components may have become small after cutting of R2.

26

For each j ∈ Z1 we find a ball B1j with H1(∂B1

j ) ≤ cθ3H1(∂Ω) and P 2j ⊂

B1j . Moreover, by Lemma 4.6 we find for each (i, j) ∈ Z2 a ball B2

i,j satisfying

H1((∂∗P 2

i ∩ ∂∗P 2j ) \ (B2

i,j ∪ Ju))

= 0 and by (30)(iv)

diam(B2i,j) ≤ cmax

k=i,jdiam(P 2

k ) maxk=i,j‖a2

i − a2j‖−1L∞(P 2

k ;R2)‖v2‖L∞(Ωgood;R2)

≤ cdiam(Ω)(Cθ−5E)−1CE ≤ cθ5H1(∂Ω),

where in the last step diam(Ω) ≤ H1(∂Ω) follows from the fact that Ω is assumedto be connected.

We define R3 =⋃j∈Z1

B1j ∪⋃

(i,j)∈Z2B2i,j and the fact that #Z1 ≤ θ−2, #Z2 ≤

I ′(I ′ − 1) ≤ θ−4 yields∑j∈Z1

H1(∂B1j ) +

∑(i,j)∈Z2

H1(∂B2i,j) ≤ cθH1(∂Ω), (34)

which together with the isoperimetric inequality gives (32)(i). We now combinedifferent components (P 2

j )I′j=1: we can find a decomposition I1∪ . . . ∪II′′ of the

indices 1, . . . , I ′ \ Z1 with the property that for each pair i1, i2 ∈ Ij, i1 < i2,we find a chain i1 = l1 < l2 < . . . , < ln = i2 such that (lk, lk+1) ∈ Z3 for allk = 1, . . . , n− 1.

Then we introduce a partition of Ωgood \ R3 consisting of the sets P 3i =⋃

j∈Ii(P2j \ R3), 1 ≤ i ≤ I ′′. (Note that this is indeed a partition of Ωgood \ R3

since, by construction, P 2j ⊂ Ωgood for j ∈ Ii and P 2

j ⊂ R3 for j ∈ Z1.) Tosee (32)(ii) we now compute using the property of the balls B1

j , B2i,j, as well as

(30)(ii), (34) and Theorem 2.7∑I′′

i=1H1(∂∗P 3

i \ J ′u) ≤ H1(∂∗Ωgood \ J ′u) +∑

j∈Z1H1(∂B1

j )

+∑

(i,j)∈Z2

(H1((∂∗P 2

i ∩ ∂∗P 2j ) \ (Ju ∪B2

i,j

)+H1(∂B2

i,j))

≤ cθH1(J ′u) + cθH1(∂Ω) + 0 ≤ cθH1(J ′u).

It remains to define v3 and to show (32)(iii). Fix (i, j) ∈ Z3. Then by the factthat (33) does not hold and mink=i,j diam(P 2

k ) ≥ θ3H1(∂Ω) ≥ θ3diam(Ω) a shortcalculation implies ‖a2

i − a2j‖L∞(Ω;R2) ≤ CE for some C = C(Ω, θ, C). Then the

triangle inequality together with #Ij ≤ I ′ ≤ θ−2 yields

maxi1,i2∈Ij ‖a2i1− a2

i2‖L∞(Ω;R2) ≤ Cθ,ΩE

for all 1 ≤ j ≤ I ′′, which by Lemma 2.3 implies maxi1,i2∈Ij |A2i1− A2

i2| ≤ Cθ,ΩE .

For each P 3j , 1 ≤ j ≤ I ′′, we choose an infinitesimal rigid motion a3

j , which coin-cides with an arbitrary a2

i , i ∈ Ij. Then (32)(iii) follows from (30)(iv).

Step IV (Conclusion). We are now in a position to prove the assertion of thetheorem. Suppose that the partition (P 3

j )I′′j=1 is ordered and choose the smallest

27

index I such that L2(P 3I+1) ≤ (θH1(J ′u))

2. Define R4 =⋃I′′

j=I+1 P3j and compute

by the isoperimetric inequality and (32)(ii)

L2(R4) ≤ θH1(J ′u)I′′∑

j=I+1

(L2(P 3j ))

12 ≤ cθH1(J ′u)

I′′∑j=I+1

H1(∂∗P 3j ) ≤ cθ(H1(J ′u))

2.

Then we define Ωbad := (Ω \Ωgood)∪ (R3 ∪R4) and by (30)(i),(ii), (32)(i),(ii) weget H1(∂∗Ωbad \ J ′u) ≤ cθH1(J ′u) and L2(Ωbad) ≤ cθ(H1(J ′u))

2.We define uθ ∈ SBV (Ω;R2) ∩ L∞(Ω;R2) by uθ = uχΩ\Ωbad

+ t0χΩbadfor

some t0 ∈ R2 such that L2(uθ 6= u4Ωbad) = 0, which is possible since u ismeasurable. Observe that the previous calculation yields (21). Let (Pi)

Ii=0 be the

Caccioppoli partition consisting of the sets P0 = Ωbad and Pi = P 3i for 1 ≤ i ≤ I.

Set ai = a3i for 1 ≤ i ≤ I and a0 = t0. Then (22)(iii) follows from (32)(iii) and

(32)(ii) yields (22)(i). Finally, the choice of the index I together with the factthat H1(J ′u) ≥ H1(∂Ω) implies (22)(ii).

Proof of Remark 4.4. Let Qλ = x+ (0, λ)2 be given and u ∈ GSBD2(Qλ). Aftertranslation we may assume x = 0. Define u ∈ GSBD2(Q1) by u(x) = u(λx) andalso note that ∇u(x) = λ∇u(λx) and H1(Ju) = λ−1H1(Ju). Applying the abovetheorem for u on Q1 we obtain uθ ∈ SBV (Q1;R2) ∩ L∞(Q1;R2) such that

(i) L2(u 6= uθ) ≤ cθ(H1(Ju) +H1(∂Q1))2,

(ii) H1((∂∗u 6= uθ ∩Q1) \ Ju) ≤ cθ(H1(Ju) +H1(∂Q1)),

a (finite) Caccioppoli partition Q1 =⋃Ii=0 Pi, and corresponding infinitesimal

rigid motions (ai)Ii=0 such that v := uθ−

∑Ii=0 aiχPi ∈ SBV (Q1;R2)∩L∞(Q1;R2)

is constant on P0 and satisfies

(i)∑I

i=0H1((∂∗Pi ∩Q1) \ Ju) ≤ cθ(H1(Ju) +H1(∂Q1)),

(ii) L2(Pi) ≥ CQ1θ2 = CQ1θ

2L2(Q1), 1 ≤ i ≤ I,

(iii) ‖v‖L∞(Q1;R2) + ‖∇v‖L1(Q1;R2×2) ≤ Cθ,Q1‖e(u)‖L2(Q1;R2×2sym).

Set Pi = λPi, uθ(x) = uθ(λ−1x) and v(x) = v(λ−1x) ∈ SBV (Qλ;R2). The

estimates for the modification in (21) follow since the estimate in (i) is two ho-mogeneous and the estimate in (ii) is one homogeneous. For the same reason(22)(i) and (23)(i) hold. We finally show (23)(ii).

Note that by transformation formula and the fact that ∇u(x) = λ∇u(λx)we have ‖e(u)‖2

L2(Qλ) = ‖e(u)‖2L2(Q1). Likewise, ‖∇v‖L1(Qλ) = λ‖∇v‖L1(Q1) and

finally we clearly have ‖v‖∞ = ‖v‖∞. Then (ii) follows as diam(Qλ) =√

2λ ≥ λ.

We now state a version of the piecewise Korn inequality with Dirichlet bound-ary conditions, which will be needed for the general existence result in Section

28

4.4, but not for the Jump Transfer Lemma. The reader more interested in thederivation of the latter may therefore wish to skip the remainder of this sectionand to proceed directly with Section 4.3.

Theorem 4.7. Let Ω ⊂ Ω′ be bounded domains in R2 with Lipschitz boundarysuch that (14) holds. Let θ > 0. Then there is a constant c = c(Ω,Ω′) > 0and some Cθ,Ω′ = Cθ,Ω′(θ,Ω

′) > 0 such that for each w ∈ H1(Ω′;R2) and u ∈GSBD2(Ω′) with u = w on Ω′ \ Ω there is a modification uθ ∈ SBV (Ω′;R2)satisfying

(i) L2(u 6= uθ) ≤ cθ(H1(Ju) + 1)2, H1(Juθ \ Ju) ≤ cθ(H1(Ju) + 1)

(ii) ‖e(uθ)‖2L2(Ω;R2×2

sym)≤ ‖e(u)‖2

L2(Ω;R2×2sym)

+ ‖∇w‖2L2(u6=uθ;R2×2),

(35)

a Caccioppoli partition Ω′ =⋃∞j=1 Pj and corresponding infinitesimal rigid mo-

tions (aj)j = (aAj ,bj)j such that v := uθ−∑∞

j=1 ajχPj ∈ SBV (Ω′;R2)∩L2(Ω′;R2),and

(i)∑∞

j=1H1((∂∗Pj ∩ Ω′) \ Ju) ≤ cθ(H1(Ju) + 1),

(ii) v = w on Ω′ \ Ω, (36)

(iii) ‖v‖L2(Ω′;R2) + ‖∇v‖L2(Ω′;R2×2) ≤ Cθ,Ω′‖e(u)‖L2(Ω′;R2×2sym) + Cθ,Ω′‖w‖H1(Ω′;R2).

As a preparation, we need the following lemma.

Lemma 4.8. Let A ⊂ R2 open, bounded with Lipschitz boundary. Then thereexists δ = δ(A) such that for all indecomposable sets E ⊂ A with finite perimetersatisfying H1(∂∗E ∩ A) ≤ δ(A) one has either

(i) L2(E) > 12L2(A) or (ii) diam(E) ≤ CAH1(∂∗E ∩ A)

for some constant CA only depending on A.

Proof. Fix ε > 0. By [9] (see also [5]) there is a constant K = K(A) and a Borelset Bε ⊂ R2 with Bε ∩ A = E such that H1(∂∗Bε) ≤ KH1(∂∗E ∩ A) + ε. It isnot restrictive to assume that Bε is indecomposable as otherwise we simply takethe component containing E. By the isoperimetric inequality we derive

minL2(Bε),L2(R2 \Bε) ≤ c(H1(∂∗Bε))2 < 1

2L2(A),

where the last inequality holds provided that δ = δ(A, c) is small enough. IfL2(R2 \Bε) <

12L2(A), we find

L2(E) = L2(Bε ∩ A) = L2(A)− L2(A \Bε) ≥ L2(A)− L2(R2 \Bε) >12L2(A)

and (i) holds. Otherwise, we particularly obtain L2(R2\Bε) = +∞ and L2(Bε) <+∞. Since Bε has finite perimeter, by an approximation argument we mayassume that Bε is bounded. As Bε is also indecomposable, Lemma 4.5 yields

diam(E) ≤ diam(Bε) ≤ H1(∂∗Bε) ≤ KH1(∂∗E ∩ A) + ε.

The claim follows with ε→ 0.

29

Proof of Theorem 4.7. By Theorem 4.3 applied with Ω′ in place of Ω we obtain aCaccioppoli partition (P ′i )

Ii=0, corresponding (a′i)

Ii=0 as well as uθ ∈ SBV (Ω′;R2)

and v′ := uθ −∑I

i=0 a′iχP ′i ∈ SBV (Ω′;R2) ∩ L∞(Ω′;R2) such that (21)-(22) hold.

Define uθ = uχΩ\P ′0 + wχP ′0 . Then (35) follows directly from (21) and (22)(ii).Let P ′ = (P ′j)

Ij=1. Let P1 ⊂ P ′ be the components completely contained in Ω

and let P2 ⊂ P ′ be the components P ′j satisfying L2(P ′j∩(Ω′\Ω)) ≥ θ. Moreover,we set P3 = P ′ \ (P1 ∪ P2). We now define the partition P = (Pj)

∞j=1 consisting

of the components

P ′0 ∪ P1 ∪ P2 ∪ P ′j ∩ Ω : P ′j ∈ P3 ∪ P ′j \ Ω : P ′j ∈ P3.

(Strictly speaking, the number of components is even finite.) For P1 := P ′0 we leta0 = 0. For Pj = P ′k ∈ P1 we set aj = a′k and for Pj = P ′k ∈ P2 we set aj = 0. IfPj ∈ P with Pj = P ′k ∩ Ω for some P ′k ∈ P3, we let aj = a′k. Finally, if Pj ∈ Pwith Pj = P ′k \ Ω for some P ′k ∈ P3, we set aj = 0.

Now define v = uθ −∑∞

j=0 ajχPj . By construction we get v = w on Ω′ \ Ω.

It remains to confirm (36)(i),(iii). To see (iii), we first note that, since uθ =v = w on the open Lipschitz set Ω′ \ Ω, by (22)(iii) with v′ in place of v and[4, Corollary 3.89], it suffices to show that the restriction of v to Ω belongs toSBV (Ω;R2) ∩ L2(Ω;R2) and that

‖v−v′‖L2(Ω;R2) +‖∇v−∇v′‖L1(Ω;R2×2) ≤ Cθ,Ω′‖e(u)‖L2(Ω′;R2×2sym) +Cθ,Ω′‖w‖H1(Ω′;R2).

(37)By construction we have that v 6= v′ ∩ Ω ⊂ (P0 ∩ Ω) ∪

⋃Pj∈P2

Pj (up to a set

of negligible measure). First, (37) with P0 ∩Ω in place of Ω follows directly from(22)(iii) and the fact that v = w on P0. Fix Pj ∈ P2. We first observe that(v′ − v)χPj = (v′ − u)χPj = a′kχPj with k such that Pj = P ′k. Since u = w on

Ω′ \ Ω we then deduce

a′kχ(Ω′\Ω)∩Pj = (v′ − w)χ(Ω′\Ω)∩Pj

and therefore

‖a′k‖L2((Ω′\Ω)∩Pj ;R2) ≤ ‖w‖L2(Ω′\Ω;R2) + ‖v′‖L2(Ω′\Ω;R2).

Consequently, using L2(Pj ∩ (Ω′ \Ω)) ≥ θ, (22)(iii) and Lemma 2.3 for ψ(s) = s2

we find|A′k|+ |b′k| ≤ Cθ,Ω′‖e(u)‖L2(Ω′;R2×2

sym) + Cθ,Ω′‖w‖L2(Ω′;R2).

Since #P2 ≤ cθ−1L2(Ω′) = C(Ω′, θ), this yields∑Pj∈P2

‖a′k‖L2(Pj ;R2) + ‖A′k‖L1(Pj ;R2×2skew) ≤ Cθ,Ω′(‖e(u)‖L2(Ω′;R2×2

sym) + ‖w‖L2(Ω′;R2)) ,

where for each j the index k is chosen such that Pj = P ′k. This implies v ∈SBV (Ω;R2) ∩ L2(Ω;R2), as well as (37), and establishes (36)(iii).

30

We now show (36)(i). To this end, we fix θ0 = θ0(Ω,Ω′) > 0 to be specifiedbelow and we first observe that it suffices to treat the case where H1(Ju) +H1(∂Ω′) ≤ θ0θ

−1. In fact, otherwise (36)(i) follows directly from (22)(i) forc = c(Ω,Ω′) large enough.

Without restriction we suppose that each P ′j ∩ (Ω′ \ Ω), P ′j ∈ P3, is indecom-posable as otherwise we consider the indecomposable components. We show thateach P ′j ∩ (Ω′ \Ω) is contained in some ball of diameter CH1(∂∗P ′j ∩ (Ω′ \Ω)) forC = C(Ω,Ω′) large enough. To see this, we first observe that due to the fact thatJu ⊂ Ω we have

H1(∂∗P ′j ∩ (Ω′ \ Ω)) ≤ cθH1(Ju) + cθH1(∂Ω′) ≤ cθ0

by (22)(i). Choose θ0 so small that cθ0 ≤ δ(Ω′ \ Ω) with δ(Ω′ \ Ω) as in Lemma4.8. Then Lemma 4.8 and the fact that L2(P ′j) ≤ θ imply for θ small

diam(∂∗P ′j ∩ (Ω′ \ Ω)) ≤ CH1(∂∗P ′j ∩ (Ω′ \ Ω)) ≤ cCθ0. (38)

We cover Θ := ∂(Ω′ \Ω) with sets U1, . . . , Un such that Ui ∩Θ is the graph ofa Lipschitz function for i = 1, . . . , n and the sets pairwise overlap such that eachball with radius cCθ0 is contained in one Ui provided that θ0 is chosen sufficientlysmall. Consequently, recalling (38) each P ′j ∩ (Ω′ \ Ω) is contained in some Ui.

Since Ui ∩ Θ is the graph of a Lipschitz function fi and P ′j ∩ (Ω′ \ Ω) ⊂⊂ Ui, itfollows that

H1(∂Ω ∩ P ′j) ≤ Lipfidiam(∂∗P ′j ∩ (Ω′ \ Ω)) ≤ CCH1(∂∗P ′j ∩ (Ω′ \ Ω)),

where C = maxi Lipfi . For the last inequality we again used (38). Finally, noting

that⋃∞j=1 ∂

∗Pj \⋃Ij=0 ∂

∗P ′j ⊂⋃P ′j∈P3

(∂Ω ∩ P ′j) we find using (22)(i)∑∞

j=1H1((∂∗Pj∩Ω′)\Ju) ≤ (1+CC)

∑I

j=0H1((∂∗P ′j∩Ω′)\Ju) ≤ cθ(H1(Ju)+1)

for c = c(Ω,Ω′) large enough.

4.3 Jump transfer lemma in GSBD

In this section we prove a jump transfer lemma which will be essential for thestability of the static equilibrium condition in the derivation of the existenceresult (Theorem 3.1).

Theorem 4.9. Let Ω ⊂ Ω′ be bounded domains in R2 with Lipschitz boundarysuch that (14) holds. Let ` ∈ N and let (wln)n ⊂ H1(Ω′;R2) be bounded sequencesfor l = 1, . . . , `. Let (uln)n be sequences in GSBD2(Ω′) and ul ∈ GSBD2(Ω′)such that

(i) ‖e(uln)‖L2(Ω′;R2×2sym) +H1(Jun) ≤M for all n ∈ N,

(ii) uln → ul in measure in Ω′, uln = wln on Ω′ \ Ω,(39)

31

for l = 1, . . . , `. Then it exists a (not relabeled) subsequence of n ∈ N with thefollowing property: For each φ ∈ GSBD2(Ω′) with H1(Jφ) < +∞ there is asequence (φn)n ⊂ GSBD2(Ω′) with φn = φ on Ω′ \ Ω such that for n→∞

(i) φn → φ in measure in Ω,

(ii) e(φn)→ e(φ) strongly in L2(Ω;R2×2sym),

(iii) H1((Jφn \

⋃`

l=1Juln) \ (Jφ \

⋃`

l=1Jul)

)→ 0.

(40)

The general strategy is to follow the proof of the SBV jump transfer (see[20, Theorem 2.1]) with the essential difference that (a) in the definition of φnwe transfer the jump not by a reflection but by a suitable extension and (b)the control of the derivatives, which is needed for the application of the coareaformula, is recovered from (39)(i) by means of the piecewise Korn inequality inTheorem 4.3.

We first concern ourselves with problem (a) and prove the following extensionlemma, based on an argument of [33].

Lemma 4.10. Let R ⊂ R2 be an open rectangle, let R− be the reflection of Rwith respect to one of its sides, and let R be the open rectangle obtained by joiningR, R− and their common side. Let φ ∈ GSBD2(R). Then it exists an extensionφ ∈ GSBD2(R) of φ satisfying

(i) H1(Jφ) ≤ cH1(Jφ)

(ii) ‖e(φ)‖L2(R;R2×2sym) ≤ c‖e(φ)‖L2(R;R2×2

sym)

(41)

for some universal constant c independent of R and φ.

Proof. We can assume R = (−l, l)×(0, h) with l, h > 0 and R− = (−l, l)×(−h, 0).For a given parameter 0 < ξ < 1 and a distribution T on (−l, l) × (0, ξh) thesymbol T ξ denotes the distribution on R− obtained by composition of T withthe diffeomorphism (x, y)→ (x,−1

ξy). We first assume φ := (φ1, φ2) is a regular

displacement in the sense of (12). Given 0 < λ < µ < 1 and p > 0 we set for all(x, y) ∈ R−

φ1(x, y) = pφ1(x,−λy) + (1− p)φ1(x,−µy)

φ2(x, y) = −λpφ2(x,−λy) + (1 + λp)φ2(x,−µy) .(42)

Furthermore φ and φ have by construction the same trace on the common bound-ary (−l, l)×0 so that no jump occurs there. With this, (41)(i) follows. In orderto show (ii), we calculate the component (Eφ)12 of the symmetrized distributionalgradient of φ. A direct computation gives

2(Eφ)12 = −λp(∂1φ2 + ∂2φ1)λ + (1 + λp)(∂1φ2)µ − µ(1− p)(∂2φ1)µ .

32

Choosing p = 1+µµ−λ gives then

2(Eφ)12 = −λp(∂1φ2 + ∂2φ1)λ + (1 + λp)(∂1φ2 + ∂2φ1)µ .

Taking the absolutely continuous parts with respect to the Lebesgue measureimplies that the L2 norm of (e(φ))12 can be controlled with the L2 norm of(e(φ))12 independently of R and φ, which was the only thing to be shown inorder to get (ii).

Before coming to the general case, we notice that the function φ has the follow-ing property: If ψ : [0,+∞) → [0,+∞) is an increasing continuous subadditivefunction satisfying (6) and

´Rψ(|φ|) dx ≤ 1, then

ˆR

ψ(|φ|) dx ≤ c (43)

again for an absolute constant c independent of R and φ. It follows indeed fromthe construction and the properties of ψ that (43) holds for a constant c onlydepending on λ, µ, and p.

In the general case φ ∈ GSBD2(R) we consider an approximating sequenceof displacements with regular jump set (φk)k in the sense of Theorem 4.2. (Againthe reader willing to assume an L2-bound may replace Theorem 4.2 by Theorem2.6.) It follows by Remark 2.2 that there exists a nonnegative concave (thus,continuous and subadditive) increasing function ψ satisfying (6) and such that

ˆR

ψ(|φk|) dx ≤ 1 .

To the functions φk we associate extensions φk ∈ GSBD2(R) satisfying (41) and(43). In particular, there is a constant C indepdendent of k such that

ˆR

ψ(|φk|) + |e(φk)|2 dx+H1(Jφk) ≤ c ,

so that (10) implies the existence of φ ∈ GSBD2(R) such that φk → φ in measurein R and

H1(Jφ) ≤ lim infk→+∞

H1(Jφk) ‖e(φ)‖L2(R;R2×2sym) ≤ lim inf

k→+∞‖e(φk)‖L2(R;R2×2

sym) . (44)

Passing to the limit and using (44), Theorem 4.2, and the corresponding inequal-ities for φk we get (41).4

4Notice that by the explicit construction of φk in (42) and the convergence in measure of φkone can also show that φ and φ have the same trace on the common boundary (−l, l)× 0.

33

We now concern ourselves with problem (b) and recall that the main as-sumption in the SBV jump transfer lemma (see [20, Theorem 2.1]) was thatthe derivatives |∇vln|, n ∈ N, are equiintegrable. Although Theorem 4.3 allowsto reconduct us to the SBV setting, we need further arguments since Theorem4.3 only provides an L1-bound for the derivatives and the bound is not given interms of the displacement field, but holds only after subtraction of a piecewiseinfinitesimal rigid motion. As a remedy, we provide the following localizationargument which is useful in the proof of Theorem 4.9, where blow up techniquesfor SBV functions are applied.

Lemma 4.11. Let θ, δ > 0 with δθ−2 ≤ 14CQ1 with the constant CQ1 from Remark

4.4. Then there is a universal constant c > 0 and some Cθ = Cθ(θ) > 0 such thatfor all squares Q ⊂ R2 and for all u ∈ GSBD2(Q), for which there are two setsB1, B2 ⊂ Q and t1, t2 ∈ R2 with

L2(Bi) ≥ (12− δ)L2(Q), ‖u− ti‖L∞(Bi;R2) ≤ δ, i = 1, 2, (45)

there is a modification uθ ∈ SBV (Q;R2)∩L∞(Q;R2) such that uθ is constant onu 6= uθ and

(i) L2(u 6= uθ) ≤ cθ(H1(Ju) + diam(Q))2,

(ii) H1(∂∗u 6= uθ \ Ju) ≤ cθ(H1(Ju) + diam(Q)),

(iii) ‖∇uθ‖L1(Q;R2×2) ≤ Cθ diam(Q)(‖e(u)‖L2(Q;R2×2

sym) + δ),

(iv) ‖uθ‖L∞(Q;R2) ≤ Cθ(‖e(u)‖L2(Q;R2×2

sym) + δ)

+ c(t1 + t2).

(46)

Proof. We apply Theorem 4.3 and obtain uθ ∈ SBV (Q;R2) ∩ L∞(Q;R2) as wellas v = uθ −

∑Ij=0 ajχPj ∈ SBV (Q;R2) for a partition (Pj)

Ij=0 and infinitesimal

rigid motions (aj)Ij=0 such that (21), (22)(i) hold. Recall that uθ is constant on

P0. Now (46)(i),(ii) follow from (21). From Remark 4.4 we get

(i) L2(Pj) ≥ CQ1L2(Q)θ2 for 1 ≤ j ≤ I,

(ii) ‖v‖L∞(Q;R2) + (diam(Q))−1‖∇v‖L1(Q;R2×2) ≤ Cθ,Q1‖e(u)‖L2(Q;R2×2sym).

(47)

By (45), (47)(i) and the assumption that δθ−2 ≤ 14CQ1 , we find L2((B1 ∪ B2) ∩

P1) ≥ L2(Pj)−L2(Q \ (B1 ∪B2)) ≥ (CQ1θ2− 2δ)L2(Q) ≥ 1

2CQ1θ

2L2(Q) for eachPj, 1 ≤ j ≤ I. Consequently, for each Pj, 1 ≤ j ≤ I, there is an index i = 1, 2such that L2(Bi∩Pj) ≥ 1

4CQ1θ

2L2(Q). Then by Lemma 2.3, (45) and (47)(ii) weget

θdiam(Q)|Aj| ≤ c‖aj − ti‖L∞(Pj∩Bi;R2) ≤ c‖v‖L∞(Pj∩Bi;R2) + c‖u− ti‖L∞(Pj∩Bi:R2)

≤ Cθ,Q1‖e(u)‖L2(Ω;R2×2sym) + cδ,

34

where c = c(CQ1) is universal. Since L2(Pj) ≤ (diam(Q))2 and #I ≤ cθ−2 by(47)(i), we calculate using (47)(ii)

‖∇uθ‖L1(Q;R2×2) = ‖∇u‖L1(Q\P0;R2×2) ≤ ‖∇v‖L1(Q;R2×2) +∑I

j=1L2(Pj)|Aj|

≤ Cθdiam(Q)(‖e(u)‖L2(Ω;R2×2

sym) + δ).

This gives (46)(iii) and finally (46)(iv) can be seen along similar lines using that‖aj − ti‖L∞(Q;R2) ≤ Cθ,Q1‖e(u)‖L2(Ω;R2×2

sym) + cδ for an index i = 1, 2.

Remark 4.12. The essential point is that (46)(iii) holds for the modification uθ,which (in contrast to the modifications in Theorem 4.3) typically coincides withthe original function outside a small set. Moreover, the estimate for ∇uθ scaleswith the diameter of the square which in the proof of Theorem 4.9 will eventuallyreplace the equiintegrability condition exploited in [20]. Later we will use that(45) can be verified for small squares whose centers are jump points of u.

We now generalize the above localization argument for a collection of squarescovering the jump set of a limiting function u = ul considered in Theorem 4.9.Moreover, we provide a modification of u which reconducts us to the setting ofscalar-valued SBV functions. In the following, for v ∈ BV and x ∈ Jv with unitnormal ν(x) we denote by Qr(x) the square with sidelength 2r, center x and twofaces perpendicular to ν(x).

Lemma 4.13. Let Ω′ ⊂ R2 open, bounded with Lipschitz boundary. Let M > 0,0 < θ, δ < 1 with δθ−8 ≤ 1

4CQ1 with the constant CQ1 from Remark 4.4. Consider

a sequence (un)n in GSBD2(Ω′) and u ∈ GSBD(Ω′) with

(i) ‖e(un)‖L2(Ω′;R2×2sym) +H1(Jun) ≤M for all n ∈ N,

(ii) un → u in measure in Ω′.(48)

Then we have for a universal constant c > 0 and Cθ = Cθ(θ) > 0 independent ofthe sequence (un)n:(a) There is v ∈ SBV (Ω′;R) ∩ L∞(Ω′;R) with H1(Ju \ Jv) ≤ θ.

(b) For all x ∈ Jv there is r(x, δ) > 0 such that, if Q∗ =⋃mi=1Qri(xi) for a finite

collection (Qri(xi))mi=1 of pairwise disjoint squares with xi ∈ Jv, ri ≤ r(xi, δ) and∑m

i=1 ri ≤M , we find a sequence (vn)n ⊂ SBV (Q∗;R) satisfying

(i) lim supn→∞

H1(Jvn \ Jun) ≤ cMθ,

(ii) lim supn→∞

‖∇vn‖L1(Q∗;R2×1) ≤ CθMδ,

(iii) lim infn→∞

‖vn − v‖L1(Qri (xi)\Fi;R) = 0 for i = 1, . . . ,m,

(49)

where Fi, i = 1, . . . ,m, are Borel sets with

35

L2(Fi) ≤ cθ2(r2i + θ2 lim infn→∞(H1(Jun ∩Qri(xi)))

2). (50)

(c) Let furthermore Ω be an open subset of Ω′ so that Ω′\Ω has Lipschitz boundary.If there exists a bounded sequence (wn)n ⊂ H1(Ω′;R2) weakly converging to w inH1(Ω′;R2) so that un = wn in Ω′ \ Ω, the sequence (vn)n can be taken such thatJvn ⊂ Ω and Jv ⊂ Ω provided the constant Cθ is allowed to additionally dependon supn ‖wn‖H1(Ω′;R2).

Proof. (a) We apply Theorem 4.1 to u and find a Caccioppoli partition Ω′ =⋃∞j=1 Pj and infinitesimal rigid motions (aj)

∞j=1 such that u −

∑j≥1 ajχPj lies in

SBV (Ω′;R2)∩L∞(Ω′;R2). Since∑+∞

j=1H1 (∂∗Pj) < +∞ and also using Theorem2.7, we choose j0 as the smallest index j such that

H1(⋃

j≥j0∂∗Pj

)+H1

(Ju ∩

⋃j≥j0

(Pj)1)≤ θ. (51)

Then we define Ωgood =⋃j0j=1 Pj and see that the function v := uχΩgood

lies in

SBV (Ω′;R2) ∩ L∞(Ω′;R2) and satisfies H1(Ju \ Jv) ≤ θ. We conclude the proofof (a) by reducing the problem to a scalar setting: denoting the two componentsof the function by v1 and v2, we can find some t ∈ (0, 1) such that the functionv = v1 + tv2 lies in SBV (Ω′;R) ∩ L∞(Ω′;R2) and satisfies

H1(Jv4Jv) = 0. (52)

This follows from the fact that At := x ∈ Jv : [v1(x)] + t[v2(x)] = 0 satisfiesH1(At) = 0 except for a countable number of t.

(b) For each x ∈ Jv let v+(x), v−(x) ∈ R2 be the one-sided limits and B±r (x) =y ∈ Qr(x) : |v(y)−v±(x)| ≤ 1

2δ for x ∈ Jv. By the definition of the approximate

limits we find r(x, δ) > 0 such that for all r ≤ r(x, δ)

L2(B+r (x)) ≥ 1

2(1− δ

)4r2, L2(B−r (x)) ≥ 1

2(1− δ)4r2. (53)

We can also additionally assume that r(x, δ) ≤ δ2.Now let Qi := Qri(xi), i = 1, . . . ,m, be a finite collection of pairwise disjoint

squares with xi ∈ Jv and ri ≤ r(xi, δ). We construct a sequence of functionsdefined on Q∗ :=

⋃mi=1Qi such that (49) holds.

For Ωgood as in (a), we define the sequence vn = unχΩgoodand note that by

(51) we obtain

H1(Jvn \ Jun) ≤ H1((∂∗Ωgood \ ∂Ω′) \ Jun

)≤ θ. (54)

Let B±i,n = y ∈ Qi : |vn(y) − v±(x)| ≤ δ and observe that due to (53) and thefact that vn → v in measure, we obtain for n large enough

L2(B+i,n) ≥

(1

2− δ)4r2, L2(B−i,n) ≥

(1

2− δ)4r2.

36

Since δθ−8 ≤ 14CQ1 , we can now apply Lemma 4.11 on vn and Qi for δ > 0 and

θ4 in place of θ to obtain a function vin ∈ SBV (Qi;R2) ∩ L∞(Qi;R2) such that(46) holds (with θ4 in place of θ).

Note that vin is constant on vin 6= vn and possibly after redefining vin we cansuppose vin = 0 on vin 6= vn. This together with the definition of vn implies (upto a set of negligible measure)

vn 6= un ⊂ vn = 0. (55)

By (46)(ii) for each i the perimeter of the sets vin 6= vn is uniformly bounded inn and by a compactness theorem for sets of finite perimeter together with (46)(i)we therefore obtain a set Fi ⊂ Qi such that χvin 6=vn → χFi in measure as n→∞,after passing to a suitable (not relabeled) subsequence. Therefore, we obtain by(46)(i) (again with θ4 in place of θ)

L2(Fi) ≤ lim infn→∞

L2(vin 6= vn) ≤ c lim infn→∞

θ4(H1(Jun ∩Qi) + diam(Qi))2,

which implies (50). As vin is bounded uniformly in L∞ (see (46)(iv)) and alsov ∈ L∞(Ω′;R2), we then deduce

lim infn→∞

‖vin − v‖L1(Qi\Fi;R2) = 0. (56)

We now pass to a scalar situation similarly as in (52). We introduce vn =∑mi=1

((vin)1 + t(vin)2

)χQi ∈ SBV (Q∗;R) and observe that (49)(iii) holds by (56).

Moreover, (46)(ii) together with∑m

i=1 ri ≤M and (54) yields (49)(i). Finally, toconfirm (49)(ii), we use (46)(iii) to compute by Holder’s inequality and the factthat ri ≤ δ2,

∑mi=1 ri ≤M

m∑i=1

‖∇vn‖L1(Qi;R2×1) ≤ cm∑i=1

‖∇vn‖L1(Qi;R2×2) ≤ Cθ

m∑i=1

ri(‖e(u)‖L2(Qi;R2×2

sym) + δ)

≤ Cθ(∑m

i=1r2i )

12‖e(u)‖L2(Ω′;R2×2

sym) + CθδM ≤ CθδM.

(c) We first notice that if wn = 0 for all n ∈ N, then the functions vn constructedabove automatically satisfy vn = 0 on Ω′ \Ω (see (55)) and thus Jvn ⊂ Ω. For vθ

we argue likewise and so the result follows. In the general case we set un = un−wnand u = u − w. Due to the boundedness of (wn)n in H1(Ω′;R2) the hypotheses(48) are satisfied for a constant also depending on supn ‖wn‖H1(Ω′;R2). We nowrepeat the proof with un in place of un and u in place of u to obtain a sequence(vn)n and v with the desired properties.

Remark 4.14. We observe that the trick in (52) can be inductively applied: ifthere are sequences (uln)n in GSBD2(Ω′) and ul ∈ GSBD2(Ω′), one can find asingle sequence (un)n converging to some u in measure and one has

H1(Jun4

⋃`

l=1Juln)

= H1(Ju4

⋃`

l=1Jul)

= 0.

37

We are now in a position to prove the jump transfer lemma.

Proof of Theorem 4.9. We first consider the case ` = 1 and drop the superscript.Let (un)n, u and φ be given as in the hypothesis. We will mainly follow theproof of the SBV jump transfer (see [20, Theorem 2.1]) employing additionallyour results obtained in Lemma 4.10 and Lemma 4.13. A further difficulty arisesfrom the fact that Lemma 4.13 is only applicable in squares where the jump setis suitably controlled (see (49)(iii), (50)) and therefore we will have to distinguishbetween good squares and bad squares.

Let θ > 0. In the following all appearing generic constants c are alwaysindependent of θ. By Lemma 4.13 (a),(c) we obtain a function v ∈ SBV (Ω′;R)∩L∞(Ω′;R) with

H1(Ju \ Jv) ≤ θ, Jv ⊂ Ω. (57)

By Et we denote the set of all Lebesgue-density 1 point for x ∈ Ω′ : v(x) > t.Let L = t ∈ R : L2(x ∈ Ω′ : v(x) = t) = 0. Then there is a countable, densesubset D ⊂ L such that Jv coincides up to a set of negligible H1-measure with

G :=⋃

t1,t2∈D,t1<t2

(∂∗Et1 ∩ ∂∗Et2 ∩ Ω′).

For x ∈ G we can choose t1(x) < t2(x) in D such that x ∈ ∂∗Et1(x) ∩ ∂∗Et2(x) andt2(x) − t1(x) ≥ 1

2|[v(x)]|, where [v(x)] denotes the jump of v. It can be shown

that ∂∗Et1(x), ∂∗Et2(x) have a common outer unit normal ν(x). Let N be the set

of points, where ∂Ω is not differentiable. We define

Gj =x ∈ G \N : |[v(x)]| ≥ 1

j, limr→0

H1((Jv \ ∂∗Et1(x)) ∩Qr(x))

2r= 0,

where Qr(x) is a square with sidelenth 2r and faces perpendicular to the normalν(x). As in the proof of [20, Theorem 2.1] we have that for fixed θ > 0 andj = j(θ) large enough

H1(Jv \Gj) ≤ θ. (58)

We also fix the half squares

Q+r (x) := y ∈ Qr(x) : (y − x) · ν(x) > 0, Q−r (x) := Qr(x) \Q+

r (x)

and the one-dimensional faces

Hr(x, s) = y ∈ Qr(x) : (y − x) · ν(x) = s, Hr(x) := Hr(x, 0).

Let δ = θ(2MjCθ)−1 with the constant Cθ from (49). Following [20, (2.3),(2.5)-

(2.6)] and covering Gj using the Morse-Besicovitch Theorem (see e.g. [19]) we

38

find a finite number of pairwise disjoint squares Qi := Qri(xi), i = 1, . . . ,m, withxi ∈ Jv, ri ≤ r(xi, δ) (cf. Lemma 4.13 (b)) such that

(i) L2(⋃m

i=1Qi

)< θ, H1(Gj \

⋃m

i=1Qi) < θ

(ii) H1((Jφ \ Jv) ∩Qi) ≤ θri,

(iii) ri ≤ H1(Jv ∩Qi) ≤ 3ri for i = 1, . . . ,m,

(iv) H1((Jv \ ∂∗Et1(xi)) ∩Qi

)≤ θri

(v) H1(y ∈ ∂∗Et1(xi) ∩Qi : dist(y,Hri(xi)) ≥ θ

2ri)≤ θri,

(vi) L2((Etk(xi) ∩Qi)4Q−i )

)≤ θ2r2

i , k = 1, 2,

(vii) Qi ⊂ Ω if xi ∈ Ω, H1(∂Ω ∩Qi) ≤ cri if xi ∈ ∂Ω,

(59)

where Q−i := Q−ri(xi). By (59)(iii) we can now apply Lemma 4.13(b) to obtain asequence (vn)n ⊂ SBV (Q∗;R) with Q∗ =

⋃mi=1 Qi satisfying (49), in particular

we have

H1(Jvn \ Jun) ≤ cMθ, Jvn ⊂ Ω. (60)

Let I ⊂ 1, . . . ,m denote the subset of good squares such that

lim infn→∞H1(Jun ∩Qi) ≤ θ−1ri (61)

if and only if i ∈ I. For the bad squares we obtain the following estimate. SinceH1(Jun ∩Qi) > θ−1ri for all i /∈ I and n ∈ N large enough, we derive by (59)(iii)∑

i/∈IH1(Jv ∩Qi) ≤

∑i/∈I

3ri ≤ 3θ∑

i/∈IH1(Jun ∩Qi) ≤ 3θM. (62)

By Ent we denote the set of all Lebesgue-density 1 point for x ∈ Ω′ : vn(x) > t.

Applying (49)(ii) we obtain recalling δ = θ(2MjCθ)−1

‖∇vn‖L1(Q∗;R2×2) ≤ CθMδ = θ2j.

In view of the coarea formula in BV this implies that there are ti ∈ [t1(xi), t2(xi)]with (see [20, (2.7)]) ∑

i∈IH1((∂∗En

ti∩Qi) \ Jvn

)≤ θ. (63)

Note that for t ∈ L, (49)(iii), (50) and (61) imply that

lim infn→∞ L2((Ent 4Et) ∩Qi) ≤ cθ2r2

i

for i ∈ I. Then taking (59)(vi) into account and following [20, (2.8)-(2.9)] wefind N(θ) such that for n ≥ N(θ)

L2((Enti∩Qi)4Q−i ) + L2((Eti ∩Qi)4Q−i ) ≤ cθ2r2

i .

39

Following [20, (2.10)-(2.14)] and using (59)(i),(iii)-(v) we get s+i , s

−i ∈ [ θ

2ri, θri]

for i ∈ I such that for n ≥ N(θ)

(i) H1(H−i \ Enti

) ≤ cθri, H1(H+i ∩ En

ti) ≤ cθri, i ∈ I,

(ii) H1(Gj \

(⋃i∈I

Ri ∪⋃

i/∈IQi

))≤ cθ,

(64)

where H+i = Hri(xi, s

+i ), H−i = Hri(xi, s

−i ) and Ri the open rectangle between

H+i and H−i .

We now transfer the jump set Jφ in Gj∩⋃i∈I Qi to

⋃i∈I(∂

∗Enti∩Qi). Assume

first xi /∈ ∂Ω. We set φ− = φχQ−i \Ri extended to Ri according to Lemma 4.10.

(This is possible when θ is small enough since Ri is a small neighborhood ofHri(xi).) In a similar way we define φ+ on (Qi \Q−i ) ∪Ri.

Now we let

φn =

φ− on Q−i \Ri,

φ+ on Qi \ (Q−i ∪Ri),

φ− on Ri ∩ Enti,

φ+ on Ri \ Enti.

If xi ∈ ∂Ω, we proceed similarly using (59)(vii) and modifying φ to φn only inthe part contained in Ω. (See again the proof of [20, Theorem 2.1] for details.Let us just mention that in this context it is crucial that the jump sets of Jv, Jvnare contained in Ω, cf. (57), (60), as hereby the function has to be indeed onlymodified in Ω.) We repeat the modification for all Qi, i ∈ I, so φn is definedon⋃i∈I Qi. Outside this union we let φn = φ. By the construction and (41) we

observe

(i) φn 6= φ ⊂(⋃

i∈IRi

)∩ Ω,

(ii) H1(Jφn ∩

⋃i∈I

Qi

)≤ cH1

(Jφ ∩

⋃i∈I

(Qi \Ri)),

(iii) ‖e(φn)‖L2(⋃i∈I Qi;R

2×2sym) ≤ c‖e(φ)‖L2(

⋃i∈I Qi;R

2×2sym).

(65)

Taking a sequence θk → 0 generates a sequence φn by choosing φn as aboveusing θk for n ∈ [N(θk), N(θk+1)). With (59)(i) and (65) we immediately deduce(40)(i),(ii).

Finally, to see (40)(iii) we again follow the argumentation in [20] and refertherein for details. By (57), (58), (62), (64)(ii) and (65)(i) we find

H1((

(Jφn \ Jun) \ (Jφ \ Ju))\(⋃

i∈IRi

))≤ O(θ).

Consequently, to conclude it suffices to show

H1(

(Jφn \ Jvn) ∩⋃

i∈IRi

)≤ O(θ), (66)

40

as the assertion the follows from (60). To this end, we consider (Jφn \ Jvn) ∩ Ri

for a fixed i ∈ I and assume xi ∈ Ω (the case xi ∈ ∂Ω is similar). We break Ri

into the parts

Ri =⋃4

k=1P ki := (Ri ∩ ∂∗En

ti) ∪ (Ri \ ∂∗En

ti) ∪ ((H+

i ∪H−i ) \ ∂∗Enti

)

∪ (∂Ri \ (H+i ∪H−i ∪ ∂∗En

ti)).

First, by (63) we have ∑i∈IH1(P 1

i \ Jvn) ≤ O(θ).

Moreover, by (58), (59)(ii),(iii), (64)(ii) and (65)(ii) we derive∑i∈I

H1(P 2i ∩ Jφn) ≤ cH1(Jφ ∩

⋃i∈I

(Qi \Ri))

≤ cH1((Jφ ∩ Jv) ∩

⋃i∈I

(Qi \Ri))

+ cH1((Jφ \ Jv) ∩⋃

iQi)

≤ O(θ).

By our construction we also have∑i∈IH1(Jφn ∩ P 3

i ) =∑

i∈I

(H1(H+

i ∩ Enti

) +H1(H−i \ Enti

))≤ O(θ),

where the last inequality follows from (59)(iii) and (64)(i). Finally, the estimate∑i∈I H1(P 4

i ) ≤ O(θ) is a consequence of (59)(iii) and |s+i |, |s−i | ≤ θri. Collecting

the previous estimates we observe that (66) holds. This concludes the proof for` = 1.

In the general case ` > 1 it suffices to apply Remark 4.14 to reduce theproblem to a single sequence (un)n for which the hypotheses of the theorem aresatisfied for a suitable bounded sequence (wn)n of boundary data.

4.4 A general compactness and existence result

Notice that for the compactness theorem in GSBD (see Theorem 2.5) it is neces-sary that the integral for some integrand ψ with limx→∞ ψ(x) =∞ is uniformlybounded. However, in many application, e.g. in our model presented below,such an a priori bound is not available. Partially following ideas in [21] we nowshow that by means of Theorem 4.7 it is possible to establish a compactness andexistence result for suitably modified functions.

We first prove the following general compactness result.

Theorem 4.15. Let Ω ⊂ Ω′ ⊂ R2 open, bounded with Lipschitz boundary suchthat (14) holds. Let w ∈ H1(Ω′,R2) and Γ be a rectifiable set with H1(Γ) < ∞.Define

E(u) =

ˆΩ

Q(e(u)) dx+H1(Ju \ Γ) (67)

41

for u ∈ GSBD2(Ω′), where Q is a positive definite quadratic form on R2×2sym.

Let M > 0 and suppose ‖w‖H1(Ω′;R2) ≤ M . Then there is an increasing concave

function ψ : [0,∞)→ [0,∞) satisfying (6) only depending on Ω,Ω′,M such thatfor every sequence (uk)k ⊂ GSBD2(Ω′) with supk≥1E(uk) ≤ M and uk = w on

Ω′\Ω we find a (not relabeled) subsequence and modifications (yk)k ⊂ GSBD2(Ω′)with yk = w on Ω′ \ Ω and

E(yk) ≤ E(uk) + 1k, supk≥1

ˆΩ′ψ(|yk|) dx ≤ 1. (68)

Moreover, there is a function y ∈ GSBD2(Ω′) with y = w on Ω′ \ Ω such that´Ω′ψ(|y|) dx ≤ 1 and for k →∞

(i) yk → y in measure on Ω′,

(ii) e(yk)→ e(y) weakly in L2(Ω′,R2×2sym),

(iii) H1(Jy \ Γ) ≤ lim infk→∞H1(Jyk \ Γ).

(69)

Note that properties (69)(ii),(iii) also hold replacing yk by uk. Moreover,observe that in general a passage to modifications is indispensable since the be-havior on components completely detached from the rest of the body cannot becontrolled.Proof. Let be given a sequence (uk)k with E(uk) ≤M and uk = w on Ω′ \Ω. As‖w‖H1(Ω′;R2) ≤M , this implies ‖e(uk)‖L2(Ω′;R2×2

sym) +H1(Juk) ≤ 2M for all k ∈ N.

Let θl = 2−2l for all l ∈ N. By Theorem 4.7 we find functions (vlk)k ⊂SBV (Ω′;R2) ∩ L2(Ω′;R2) of the form

vlk = ulk −∑∞

j=1ak,lj χPk,lj

, (70)

where ulk are modifications, (P k,lj )j are partitions of Ω′ and (ak,lj )j infinitesimal

rigid motions. In particular, for all l ∈ N, k ∈ N we have vlk = w on Ω′ \ Ω, themodifications satisfy (see (35))

(i) H1(Julk \ Juk) ≤ cθl, (ii) ‖e(ulk)‖2L2(Ω;R2×2

sym)≤ ‖e(uk)‖2

L2(Ω;R2×2sym)

+ εl. (71)

for some c = c(M,Ω,Ω′) > 0, where (εl)l is a null sequence only depending on w,and

(i) ‖vlk‖L2(Ω′;R2) ≤ Cl, (ii) ‖e(vlk)‖L2(Ω′,R2×2sym) ≤ 3M, (iii) H1(Jvlk \ Juk) ≤ cθl

(72)

for Cl = Cl(θl, w,M,Ω′) > 0 (cf. (36)). Here we used, possibly passing to alarger M , that εl ≤ M for all l ∈ N. Without restriction we can assume that Clis increasing in l.

42

Using a diagonal argument we get a (not relabeled) subsequence of (k)k∈Nsuch that by Theorem 2.5 for every l ∈ N we find a function vl ∈ GSBD2(Ω′)with vlk → vl in L1(Ω′;R2) for k →∞ and

e(ulk) = e(vlk) e(vl) weakly in L2(Ω′;R2×2sym), H1(Jvl \ Γ) ≤ lim inf

k→∞H1(Jvlk \ Γ).

In particular, by (72) we have

‖vl‖L1(Ω′;R2) ≤ c‖vl‖L2(Ω′;R2) ≤ Cl, ‖e(vl)‖L2(Ω′;R2×2sym) +H1(Jvl) ≤ 4M + c. (73)

Likewise, we can establish a compactness result for the Caccioppoli partitions.By construction (see (70)) and (72)(iii) we have∑

jH1(∂∗P k,l

j ∩ Ω′) ≤ 2H1(Juk ∪ Jvlk) ≤ 2M + 2c (74)

for all k, l ∈ N. Thus, by Theorem 2.8 we find for all l ∈ N an (ordered) partition(P l

j)j with∑

jH1(∂∗P lj ∩Ω′) ≤ 2M + 2c such that for a suitable subsequence one

has P k,lj → P l

j in measure for all j ∈ N as k → ∞ and∑

j L2(P k,lj 4P l

j

)→ 0

for k → ∞. As∑

jH1(∂∗P lj ∩ Ω′) ≤ 2M + 2c for all l ∈ N, we can repeat the

arguments and obtain a partition (Pj)j such that∑

j L2(P lj4Pj

)→ 0 for l→∞

after extracting a suitable subsequence. Consequently, using a diagonal argumentwe can choose a (not relabeled) subsequence of (l)l∈N and afterwards of (k)k∈Nsuch that∑

jL2(P lj4Pj

)≤ 2−l,

∑jL2(P k,lj 4P l

j

)≤ 2−l for all k ≥ l. (75)

We now want to pass to the limit l → ∞ for the sequence (vl)l. However, wesee that the compactness result in GSBD cannot be applied directly as the L1

bound depends on θl (cf. (73)). We show that by choosing the infinitesimalrigid motions on the elements of the partitions appropriately (see (70)) we canconstruct the sequence (vl)l such that

L2(⋂

m≥n|vn − vm| ≥ 1

)≤ c2−n for all n ∈ N (76)

for a constant c = c(M,Ω,Ω′) > 0, whence Lemma 2.1 is applicable.We fix k ∈ N and describe an iterative procedure to redefine ak,lj = aAk,lj ,bk,lj

for

all l, j ∈ N. Let v1k = v1

k as defined in (70) and assume vlk as well as (ak,lj )j have

been chosen (which may differ from (ak,lj )j) such that (72)(i) still holds possibly

passing to a larger constant Cl. Fix some P k,l+1j , j ∈ N. If L2(P k,l

j ∩Pk,l+1j ) > θl+1,

we define ak,l+1j = ak,lj on P k,l+1

j . Otherwise, we set ak,l+1j = ak,l+1

j . In the firstcase we then obtain by the triangle inequality

‖ak,l+1j −ak,l+1

j ‖L1(Pk,lj ∩Pk,l+1j ;R2) ≤ ‖v

lk‖L1(Ω′;R2)+‖vl+1

k ‖L1(Ω′;R2) ≤ Cl+Cl+1 ≤ 2Cl+1.

43

In the penultimate step we have used that (72)(i) holds for vlk and vl+1k . Conse-

quently, by Lemma 2.3 for ψ(s) = s we find |bk,l+1j − bk,l+1

j | ≤ C l+1∗ for a constant

C l+1∗ only depending on Ω′, Cl+1, θl+1 and M . We define vl+1

k as in (70) replacing

ak,l+1j by ak,l+1

j and summing over all components we derive

‖vl+1k ‖L1(Ω′;R2) ≤ ‖vl+1

k ‖L1(Ω′;R2) + C l+1∗ L2 (Ω′) ≤ Cl+1 + C l+1

∗ L2 (Ω′) ,

i.e. (72)(i) is also satisfied for vl+1k after possibly passing to a larger constant

Cl+1 = Cl+1(θl+1,M,Ω′).For simplicity the modified functions and the infinitesimal rigid motions will

still be denoted by vlk and ak,lj in the following. We now show that (76) holds. Tothis end, we define Ank,l =

⋂n≤m≤lvmk = vnk for all n ∈ N and n ≤ l ≤ k. If we

show

L2(Ω′ \ Ank,l

)≤ c2−n, (77)

then (76) follows. Indeed, for given l ≥ n we can choose K = K(l) ≥ l so largethat L2

(|vmK − vm| > 1

2)≤ 2−m for all n ≤ m ≤ l since vmk → vm in L1(Ω′;R2)

for k →∞. This implies

L2(⋂

n≤m≤l|vm − vn| ≥ 1

)≤ L2

(Ω′ \ AnK,l

)+∑

n≤m≤lL2(|vmK − vm| > 1

2)≤ c2−n.

Passing to the limit l→∞ we then derive L2(⋂

m≥n|vm − vn| ≥ 1)≤ c2−n, as

desired.We now confirm (77). To this end, fix k ≥ l and observe that Ank,l =⋂

n≤m≤lT nk = Tmk , where T nk =∑

j ak,nj χPk,nj

. We consider Tmk = Tm+1k

for n ≤ m ≤ l − 1 and from (75) we deduce∑

j L2(P k,m+1j 4P k,m

j

)≤ 3 · 2−m.

Define J1 ⊂ N such that L2(P k,m+1j

)≤ 2θm+1 for j ∈ J1. Then let J2 ⊂ N \ J1

such that L2(P k,m+1j ∩ P k,m

j

)> 1

2L2(P k,m+1j

)for all j ∈ J2. Finally, we observe

that L2(P k,m+1j

)≤ 2L2

(P k,m+1j \ P k,m

j

)for j ∈ J3 := N \ (J1 ∪ J2). Using the

isoperimetric inequality and (74) we derive∑j∈J1

L2(P k,m+1j

)≤√

2√θm+1

∑j∈J1

L2(P k,m+1j

) 12

≤ c2−m∑j∈J1

H1(∂∗P k,m+1j ) ≤ c(M + c)2−m.

44

Due to the above construction of the infinitesimal rigid motions we obtain Tmk =Tm+1k ⊃

⋃j∈J2(P

k,m+1j ∩ P k,m

j ) and therefore

L2(Ω′ \ Tmk = Tm+1

k )≤∑j∈J2

L2(P k,m+1j \ P k,m

j

)+

∑j∈J1∪J3

L2(P k,m+1j

)≤∑j∈J2

L2(P k,m+1j \ P k,m

j

)+∑j∈J3

2L2(P k,m+1j \ P k,m

j

)+ c(M + c)2−m ≤ c2−m

for c only depending on M,Ω,Ω′. Summing over n ≤ m ≤ l− 1 we establish (77)and consequently (76).

In view of (73) and (76) we can apply Lemma 2.1 on the sequences sl = Cland tl = c2−l to obtain an increasing, concave function ψ with (6) such thatsupl≥1

´Ω′ψ(|vl|) dx ≤ 1. Define ψ(s) = 1

2minψ(s), s and observe that ψ has

the desired properties. In particular, the choice of ψ only depends on Ω,Ω′ andM .Recalling vlk → vl in L1(Ω′;R2), (71) and (72)(iii) we can now select a subsequenceof (uk)k and a diagonal sequence (yk) ⊂ (vlk)k,l such that ‖yk − vl‖L1(Ω′;R2) ≤ 1for some vl and E(yk) ≤ E(uk) + 1

k. Then we get that (68) holds.

The existence of a function y ∈ GSBD2(Ω′) with y = w on Ω′ \ Ω and´Ω′ψ(|y|) dx ≤ 1 as well as the convergence (69) now directly follow from Theorem

2.5.

As a consequence we now obtain the following existence result.

Theorem 4.16. Let Ω ⊂ Ω′ ⊂ R2 open, bounded with Lipschitz boundary suchthat (14) holds. Let w ∈ H1(Ω′,R2) with ‖w‖H1(Ω′;R2) ≤ M and E as given in(67). Then the following holds:(i) There is a minimizer of E(u) among all functions u ∈ GSBD2(Ω′) with u = won Ω′ \ Ω.(ii) There is an increasing concave function ψ : [0,∞) → [0,∞) with (6) onlydepending on Ω,Ω′,M such that

´Ω′ψ(|u|) dx ≤ 1 for at least a minimizer u of

the minimization problem in (i).

Proof. Let A := u ∈ GSBD(Ω′) : u = w on Ω′ \ Ω and (uk)k ⊂ A withE(uk)→ infu∈AE(u). We employ Theorem 4.15 and let (yk)k be a (sub-)sequenceof modifications converging to u ∈ A in the sense of (69). Then we find by(68),(69)

E(u) ≤ lim infk→∞E(yk) = lim infk→∞E(uk) = infu∈AE(u).

Consequently, u is a minmizer of the minimization problem in (i). Moreover,by Theorem 4.15 we find a function ψ with the desired properties such that´

Ω′ψ(|u|) ≤ 1.

We later will use property (ii) to derive compactness in GSBD2 of the mini-mizers of our incremental problems. Concerning the stability of minimizers with

45

respect to converging sequences of boundary data we have the following corollarybeing a consequence of the Jump Transfer Lemma. As before Q is a strictlypositive quadratic form on R2×2

sym.

Corollary 4.17. Let Ω ⊂ Ω′ ⊂ R2 open, bounded with Lipschitz boundary suchthat (14) holds. Let Γ ⊂ R2 be a measurable set with H1(Γ) < ∞, let (un)n, u ∈GSBD2(Ω′) with H1(Ju) < ∞ and un = wn in Ω′ \ Ω for (wn)n ⊂ H1(Ω′;R2)such that d(un, u)→ 0, e(un) e(u) weakly in L2(Ω′;R2×2

sym). If un minimize

ˆΩ

Q(e(v)) +H1(Jv \ (Jun ∪ Γ))

among all functions with the same Dirichlet data, then u minimizes

ˆΩ

Q(e(v)) +H1(Jv \ (Ju ∪ Γ))

among all functions v such that v = u on Ω′ \ Ω. If furthermore (wn)n is aconstant sequence, we have e(un)→ e(u) strongly in L2(Ω′;R2×2

sym).

The proof is omitted as it is completely analogous to Corollary 2.10 in [20]provided one substitutes the Dirichlet energy with the linearized elastic energyand gradient by the symmetrized gradient.

5 Proof of the existence result

Equipped with the theoretical results in the previous section, we can obtain theannounced existence result Theorem 3.1 by passing to the limit in the usualscheme of time-incremental minimization. The discussion in this section willclosely follow the analogous one in [20, Section 3] and therefore not all the proofswill be detailed. For the reader’s convenience we will only focus on some pointswhere our GSBD2 setting involves some modifications of the arguments devel-oped there. Through all this section we will write H1(Γ) in place of H1(Γ ∩ Ω′),since all the cracks we consider in the proof will have by construction no inter-section with ∂Ω \ ∂DΩ.

We fix a time interval [0, T ] and consider a countable dense subset I∞ thereof.We can assume that 0 and T belong to I∞. For each n ∈ N we choose a subsetIn := 0 = tn0 < tn1 < · · · < tnn = T such that (In)n form an increasing sequence ofnested sets whose union is I∞. Setting ∆n := sup

1≤k≤n(tnk−tnk−1), we have that ∆n →

0 when n→ +∞. As discussed in Section 3, we consider a boundary datum g ∈W 1,1([0, T ];H1(R2;R2)) and the corresponding left-continuous piecewise constantinterpolation

gn(t) := g(tnk) for all t ∈ [tnk , tnk+1)

46

which satisfies g(t) = gn(t) for all t ∈ I∞, when n is large enough. Moreover,gn(t) → g(t) strongly in H1 for all t ∈ [0, T ]. We set un(0) = u(0), the giveninitial datum, while for all k = 1, . . . , n we recursively define unk as a minimizerof the problem

ˆΩ

Q(e(v)) dx+H1

(Jv \

⋃0≤j≤k−1

Junj

)(78)

among the functions v ∈ GSBD2(Ω′) satisfying v = g(tnk) in Ω′\Ω. The existenceof such a minimizer follows from Theorem 4.16. We then construct left-continuouspiecewise constant interpolation

un(t) := unk for all t ∈ [tnk , tnk+1) .

The following a-priori estimates on the interpolations can be then derived com-bining similar arguments as those developed in [16] and [20] with the additionalproperty (ii) of Theorem 4.16.

Lemma 5.1. There exists an increasing concave function ψ : [0,∞) → [0,∞),satisfying (6), which only depends on Ω,Ω′ and supt∈[0,T ] ‖g(t)‖H1, such that theinterpolations un(t) satisfy

ˆΩ′ψ(|un(t)|) dx+ ‖e(un(t))‖L2(Ω′;R2×2

sym) +H1

( ⋃τ∈I∞ , τ≤t

Jun(τ)

)≤M (79)

for a constant M independent of t ∈ [0, T ]. Furthermore, setting σn(t) :=Ce(un(t)) with C as in (13), it exists a modulus of continuity ω such that thefollowing energy inequality holds at every t ∈ [0, T ]:

ˆΩ

Q(e(un(t)) dx+H1

( ⋃τ∈I∞ , τ≤t

Jun(τ)

)

≤ˆ

Ω

Q(e(u(0))) dx+H1(Ju(0)

)+

ˆ t

0

〈σn(s), e(g(s))〉 ds+ ω(∆n) . (80)

Proof. The bound on ‖e(un(t))‖L2(Ω′,R2×2sym) is simply obtained by comparing the

minimizer un(t) with the admissible competitor gn(t), while the existence of a ψas in (79) follows from (ii) in Theorem 4.16 and the assumptions on g. Fix nowt ∈ [0, T ], and for fixed n, let k be such that t ∈ [tnk , t

nk+1). By construction, since

In ⊂ I∞, one has ⋃τ∈I∞,τ≤t

Jun(τ) =k⋃j=0

Junj .

Testing for every 1 ≤ j ≤ k the minimality of un(tnj ) with the admissible com-petitor un(tnj−1)+g(tnj )−g(tnj−1), summing up all steps until step k and using the

47

above equality, we obtain (80) (for the details, use the same arguments leadingto [20, (3.4)], upon replacing the Dirichlet energy with the linearized elastic en-

ergy.) Once (80) is proved, the uniform a-priori bound on H1(⋃

τ∈I∞,τ≤t Jun(τ)

)simply follows by the Cauchy-Schwarz inequality and the already proven boundon σn(t).

The following lower semicontinuity result will be needed in order to pass tothe limit in the previous bounds. We do not report the proof, which is verbatimthe same as in [20, Lemma 3.1], provided one uses the GSBD compactness andlower semicontinuity theorem in place of the one in SBV .

Lemma 5.2. Let A ⊂ R2 be open, bounded. For all ` ∈ N, let (vn` )n be asequence of functions in GSBD(A) satisfying the assumptions of Theorem 2.5,and let v` ∈ GSBD(A) be such that d(vn` , v`)→ 0 when n→ +∞. Then

H1

(+∞⋃`=0

Jv`

)≤ lim inf

n→+∞H1

(+∞⋃`=0

Jvn`

).

Using the bounds in (79), we will initially define u(t) only for t ∈ I∞. Thiswill already allow us to define a crack set Γ(t) for all t ∈ [0, T ] with Ju(t) ⊂ Γ(t)for t ∈ I∞. The function u(t) will be later extended to all t in a way that theinclusion Ju(t) ⊂ Γ(t) still holds.

Theorem 5.3. There exists a (not relabeled) subsequence (un(t))n independentlyof t ∈ I∞ and a function u : I∞ → GSBD2(Ω′) such that d(un(t), u(t)) → 0 forall t ∈ I∞ and, setting

Γ(t) :=⋃

τ∈I∞ , τ≤t

Ju(τ) for all t ∈ [0, T ] , (81)

the following properties are satisfied:

(i) u(t) = g(t) in Ω′ \ Ω for all t ∈ I∞,(ii) e(un(t))→ e(u(t)) strongly in L2(Ω′,R2×2

sym) for all t ∈ I∞,

(iii) H1(Γ(t)) ≤ lim infn→∞

H1

( ⋃τ∈I∞ , τ≤t

Jun(τ)

)for all t ∈ [0, T ] .

(82)

Furthermore, for all t ∈ I∞, u(t) minimizes

ˆΩ

Q(e(v)) +H1(Jv \ Γ(t)) (83)

among all functions v such that v = g(t) on Ω′ \ Ω.

48

Proof. By (79) the sequence (un(t))n satisfies the assumptions of Theorem 2.5for every t ∈ I∞. With this, up to extracting a diagonal sequence, there existsu : I∞ → GSBD2(Ω′) such that d(un(t), u(t))→ 0 and e(un(t)) e(u(t)) weaklyin L2(Ω′,R2×2

sym) for all t ∈ I∞. Since un(t) = gn(t) in Ω′ \ Ω and gn(t) = g(t) forn large enough, (82)(i) follows. At the expense of a numbering of I∞, (82)(iii)follows from Lemma 5.2.

From the definition of un(t) and gn(t) (cf. (78)), for all t ∈ [0, T ] we have thatun(t) is minimizing

ˆΩ

Q(e(v)) dx+H1

(Jv \

⋃τ∈In , τ≤t

Jun(τ)

)(84)

among the functions v ∈ GSBD2(Ω′) satisfying v = gn(t) in Ω′ \ Ω. A fortiori,we deduce that un(t) is a minimizer with respect to its own jump set, that iswith Jun(t) in place of

⋃τ∈Inτ≤t

Jun(τ) in the above problem. If additionally t ∈ I∞,

we can choose n so large that t ∈ In ∩ I∞, and thus gn(t) = g(t). With this,Corollary 4.17 gives (82)(ii).

We now fix δ > 0 and t ∈ I∞. Since H1(Γ(t)) is finite, we can find ` ∈ N sothat t ∈ I` and the subset Γ`(t) of Γ(t) defined by

Γ`(t) =⋃

τ∈I` , τ≤t

Ju(τ)

satisfiesH1(Γ(t)\Γ`(t)) < δ. For all n ≥ `, we similarly define Γn` (t) with un(τ) inplace of u(τ). Notice that Ju(t) ⊂ Γ`(t) and Jun(t) ⊂ Γn` (t) since t ∈ I`. With thisand using (84) we have that un(t) is minimizing

´ΩQ(e(v)) dx +H1 (Jv \ Γn` (t))

among the functions v ∈ GSBD2(Ω′) which satisfy v = g(t) in Ω′ \ Ω.We observe that by Lemma 5.1 the sequences (un(τ))n with τ ∈ I`, τ ≤ t,

and the corresponding limiting functions u(τ) defined above satisfy (39). Conse-quently, for any v with v = g(t) in Ω′\Ω we can apply Theorem 4.9 to φ = v−u(t)and to the finite unions of jump sets Γn` (t) and Γ`(t). Therefore, we get the ex-istence of a sequence (φn)n such that φn = v − u(t) = 0 in Ω′ \ Ω satisfying, by(40) and (82)(ii),

‖e(un(t) + φn)− e(v)‖L2(Ω′,R2×2sym) → 0, lim sup

n→+∞H1(Jφn \ Γn` (t)) ≤ H1(Jφ \ Γ`(t))

(85)

as n → +∞. Furthermore, since t ∈ I∞, when n is so big that gn(t) = g(t) inΩ′ \Ω we have that un(t) +φn = g(t) in Ω′ \Ω. The minimality of un(t), (82)(ii),

49

and (85) then imply thatˆ

Ω

Q(e(u(t))) dx = limn+∞

ˆΩ

Q(e(un(t))) dx

≤ lim supn→+∞

ˆΩ

Q(e(un(t) + φn)) dx+H1(Jun(t)+φn \ Γn` (t)

)≤ˆ

Ω

Q(e(v)) dx+H1 (Jv \ Γ`(t)) ≤ˆ

Ω

Q(e(v)) dx+H1 (Jv \ Γ(t)) + δ ,

where we in the third step we used that Ju(t) ⊂ Γ`(t) and Jun(t) ⊂ Γn` (t). Thisfinishes the proof of (83) since δ is arbitrary.

Remark 5.4. Let t /∈ I∞ and let w ∈ GSBD2(Ω′) be such that (un(t))n hasa subsequence, possibly depending on t, which converges to w in the sense of(10). Fix δ > 0 and Γ`(t) and Γn` (t) as in the previous proof, without the requestt ∈ I`. We can apply Theorem 4.9 for the finite number of sequences (un(t))nand (un(τ))n with τ ∈ I`, τ ≤ t, and thus for any v with v = g(t) in Ω′ \ Ω, wecan apply (40) to φ = v to obtain a corresponding sequence (φn)n. It follows nowfrom (84) that (with vn := φn + gn(t)− g(t))

ˆΩ

Q(e(un(t))) dx ≤ˆ

Ω

Q(e(vn)) dx+H1(Jvn \(Γn` (t) ∪ Jun(t)

)) .

By (10)(ii), the strong convergence of gn(t) to g(t) in H1, (40) and the arbitrari-ness of δ we deduce the minimality propertyˆ

Ω

Q(e(w)) dx ≤ˆ

Ω

Q(e(v)) dx+H1(Jv \ (Γ(t) ∪ Jw)) . (86)

For v = w one also gets limn→+∞

ˆΩ

Q(e(un(t))) dx =

ˆΩ

Q(e(w)) dx, which implies

‖e(un(t))− e(w)‖L2(Ω′,R2×2sym) → 0 (87)

by the strict convexity of Q.

In the next theorem we extend u from I∞ to a function defined on all of[0, T ]. We prove that this extension satisfies the inclusion Ju(t) ⊂ Γ(t) for allt ∈ [0, T ] (notice that, at this point of the stage, the crack set Γ(t) is alreadydefined on the whole interval [0, T ]), the global minimality condition, as well asthe “≤”-inequality in the energy balance of Theorem 3.1. The proof follows veryclosely in the footsteps of [20, Lemma 3.8]: A sketch is reported for the reader’sconvenience.

Theorem 5.5. There exists a function u : [0, T ]→ GSBD2(Ω′) with u(t) = g(t)in Ω′ \ Ω and an H1-rectifiable crack Γ(t) ⊂ Ω, nondecreasing in t, such thatJu(t) ⊂ Γ(t) for all t ∈ [0, T ] and:

50

• (global stability) for all t ∈ [0, T ], u(t) minimizesˆ

Ω

Q(e(v)) dx+H1 (Jv \ Γ(t))

among the functions v ∈ GSBD2(Ω′) which satisfy v = g(t) in Ω′ \ Ω.

• (energy inequality) defining the stress σ(t) and the total energy E(t) as inTheorem 3.1, it holds

E(t) ≤ E(0) +

ˆ t

0

〈σ(s), e(g(s))〉 ds .

Proof. We consider u : I∞ → GSBD2(Ω′) as in Theorem 5.3. Accordingly, wedefine Γ(t) as in (81) for all t ∈ [0, T ]. Thus, we simply have to define u when t /∈I∞. We fix t /∈ I∞ and an increasing sequence (tk)k ⊂ I∞ converging to t. Noticethat for the interpolants un(t) the inequality (79) holds with a constant M anda function ψ which are not depending on k. Since, for all k, d(un(tk), u(tk))→ 0when n→ +∞ and thus also un(tk)→ u(tk) a.e. for a not relabeled subsequence,by Fatou’s lemma and (10), also the sequence (u(tk))k satisfies (79). Then, itexists a limit point u(t) ∈ GSBD2(Ω′) with d(u(tk), u(t)) → 0 and e(u(tk)) e(u(t)) weakly in L2 as k → ∞. It is obvious that, u(t) = g(t) in Ω′ \ Ω whilean application of (83) together with the arguments leading to [20, (3.24)], againsimply using GSBD in place of SBV compactness, shows that the inclusionJu(t) ⊂ Γ(t) holds up to a H1-null set.

We now prove the global stability property. Notice that for all k one has bydefinition Γ(tk) ⊂ Γ(t) and, since the sequence of cracks Γ(tk) is nondecreasing,it holds that H1(Γ(t) \ Γ(tk)) → 0 when k → +∞. For each v ∈ GSBD2(Ω′)with v = g(t) in Ω′ \Ω, the sequence vk = v+ g(tk)− g(t) has the same jump setas v and clearly satisfies e(vk)→ e(v) in L2(Ω′,R2×2

sym). By Theorem 5.3 we have

ˆΩ

Q(e(u(tk))) dx ≤ˆ

Ω

Q(e(vk)) dx+H1(Jv \ Γ(tk)) .

Taking the limit we get the global stability because of the inclusion Ju(t) ⊂ Γ(t).We also get, for v = u(t) in the above argument, that

limn→+∞

ˆΩ

Q(e(un(t))) dx =

ˆΩ

Q(e(u(t))) dx,

which implies the strong convergence of e(u(tk)) to e(u(t)). Furthermore, due tothe strict convexity of Q, the function e(u(t)) is uniquely determined by the globalstability and the condition Ju(t) ⊂ Γ(t). Thus, e(u(t)) is uniquely determinedonce I∞ is fixed. This implies the strong convergence of e(u(tk)) to e(u(t)) on thewhole sequence (tk)k and not only along a subsequence, and that the mappingt→ e(u(t)) is strongly left continuous in L2 at any t ∈ [0, T ] \ I∞.

51

The energy inequality immediately follows from (80) and (82)(iii), once thefollowing claim is proved:

e(un(t))→ e(u(t)) strongly in L2(Ω′,R2×2sym)

for a.e. t ∈ [0, T ]. In fact, one can then pass to the limit in (80) also in the termassociated to the work of the external loads. Because of (82), it suffices to showthe claim for t ∈ [0, T ] \ I∞. Notice that because of (79) the L2-norm of thesequence e(un(t)) is bounded. Furthermore, again (79), together with Theorem2.5 imply that any weak accumulation point of e(un(t)) must be of the form e(w),where w is a GSBD2 function such that a subsequence, possibly depending ont, of un(t) converges to w in the sense of (10). Therefore, to prove the claim itsuffices to show that e(w) = e(u(t)) for a.e. t, so that the limit is independent ofthe chosen subsequence and the strong convergence holds because of (87).

Let us a consider a weak accumulation point w. Notice that u(t) is an admis-sible competitor for the problem (86), which as shown above additionally satisfiesJu(t) ⊂ Γ(t). Therefore, if we prove

ˆΩ

Q(e(u(t))) dx ≤ˆ

Ω

Q(e(w)) dx (88)

for a.e. t, we will get e(w) = e(u(t)) as requested, otherwise, using the strictconvexity of Q, we would contradict (86) with v = 1

2(w + u(t)). Now, using (87)

and the left continuity of t→ e(u(t)) at t /∈ I∞, the inequality (88) follows fromthe minimality of un(t) arguing exactly as in the proof of part (d) in [26, Lemma4.3], again upon substituting the Dirichlet with the linear elastic energy. We omitthe details.

We are finally in a position to give the proof of Theorem 3.1.

Proof of Theorem 3.1. Defining u(t) as in Theorems 5.3 and 5.5 and Γ(t) as in(81), the only thing left to show is the “≥”-inequality in the energy balance. Thisfollows from global stability by a well-known argument (see [26, Lemma 4.6]) thatwe sketch for the reader’s convenience. We first notice that the map t 7→ H1 (Γ(t))is bounded monotone increasing, so that it is continuous at each t ∈ [0, T ] \ N ,where N has 0-Lebesgue measure. At each t ∈ [0, T ] \ (I∞ ∪N ) we already nowthat e(u(·)) is left continuous with respect to to the L2-norm. We can show thatit is indeed continuous, arguing as follows. Fixing a decreasing sequence tk → t,any weak-L2 accumulation point e(w) of e(u(tk)) satisfies, because of (10), theinclusion Γ(t) ⊂ Γ(tk), and the continuity of H1 (Γ(·)) at time t, that

H1 (Jw \ Γ(t)) ≤ lim infk→+∞

H1(Ju(tk) \ Γ(tk)

)= 0 .

Consequently, Jw ⊂ Γ(t) up to a H1-null set. With this, testing the globalstability of u(tk) with v + g(tk) − g(t) for any v with v = g(t) in Ω′ \ Ω and

52

arguing as in the proof of Theorem 5.5, we obtain e(w) = e(u(t)) as well as theclaimed strong convergence.

Fix now t ∈ [0, T ]. Setting for every k ∈ N and every i = 0, . . . , k, sik = ikt

and uk(s) = u(si+1k ) whenever t ∈ (sik, s

i+1k ], we have that ‖e(uk(s))‖L2(Ω′,R2×2

sym) is

uniformly bounded because of the energy inequality (see Theorem 5.5), and

‖e(uk(s))− e(u(s))‖L2(Ω′,R2×2sym) → 0 for all s ∈ [0, t] \ (I∞ ∪N ) , (89)

that is a.e. in [0, t]. Testing the global stability of u(sik) with u(si+1k )− g(si+1

k ) +g(sik), summing up on i and exploiting the absolute continuity of t 7→ g(t), oneobtains

E(t) ≥ E(0) +

ˆ t

0

〈σk(s), e(g(s))〉 ds+ ηk ,

where σk(s) := Ce(uk(s)) and ηk is an infinitesimal remainder. The thesis nowfollows by dominated convergence and (89) when taking the limit k → +∞.

Acknowledgements This work has been funded by the Vienna Science andTechnology Fund (WWTF) through Project MA14-009.

References

[1] L. Ambrosio. A compactness theorem for a new class of functions ofbounded variation. Boll. Un. Mat. Ital. B(7) 3 (1989), 857–881.

[2] L. Ambrosio. Existence theory for a new class of variational problems.Arch. Ration. Mech. Anal. 111 (1990), 291–322.

[3] L. Ambrosio, A. Coscia, G. Dal Maso. Fine properties of functionswith bounded deformation. Arch. Ration. Mech. Anal. 139 (1997), 201–238.

[4] L. Ambrosio, N. Fusco, D. Pallara. Functions of bounded variationand free discontinuity problems. Oxford University Press, Oxford 2000.

[5] A. Baldi, F. Montefalcone. A note on the extension of BV functionsin metric measure spaces. J. Math. Anal. Appl. 340 (2008), 197–208.

[6] G. Bellettini, A. Coscia, G. Dal Maso. Compactness and lowersemicontinuity properties in SBD(Ω). Mathematische Zeitschrift 228 (1998),337–351.

[7] A. Braides, A. Defranceschi, E. Vitali. Homogenization of free dis-continuity problems. Arch. Ration. Mech. Anal. 135 (1996), 297–356.

[8] D. Bucur, N. Varchon. Boundary variation for a Neumann problem.Ann. Scuola Norm. Sup. Cl. Sci. 29(4) (2000), 807–821.

53

[9] Yu. D. Burago, V. G. Maz’ja. Potential Theory and Function Theoryfor irregular regions. Translated from Russian. Seminars in Mathematics, V.A. Steklov Math. Ins., Leningrad, vol. 3, 1969.

[10] A. Chambolle. A density result in two-dimensional linearized elasticity,and applications. Arch. Ration. Mech. Anal. 167 (2003), 211–233.

[11] A. Chambolle. An approximation result for special functions with boundeddeformation. J. Math. Pures Appl. 83 (2004), 929–954.

[12] A. Chambolle, F. Doveri. Continuity of Neumann linear elliptic prob-lems on varying two-dimensional bounded open sets. Comm. Partial Differ-ential Equations 22 (1997), 811–840.

[13] S. Conti, M. Focardi, F. Iurlano. Integral representation for func-tionals defined on SBDp in dimension two. preprint ArXiv: 1510.00145,(2015).

[14] G. Dal Maso. Generalised functions of bounded deformation. J. Eur. Math.Soc. (JEMS) 15 (2013), 1943–1997.

[15] G. Dal Maso, G. A. Francfort, R. Toader. Quasistatic crack growthin nonlinear elasticity. Arch. Ration. Mech. Anal. 176 (2005), 165–225.

[16] G. Dal Maso, R. Toader. A model for the quasi-static growth of brittlefractures: existence and approximation results. Arch. Ration. Mech. Anal.162 (2002), 101–135.

[17] E. De Giorgi. Free-discontinuity problems in calculus of variations. InR. Dautray, editor, Frontiers in pure and applied mathematics, a collectionof papers dedicated to J.-L. Lions on the occasion of his 60th birthday, pages55–62. North Holland, 1991.

[18] E. De Giorgi, M. Carriero, A. Leaci. Existence theorem for a min-imum problem with free discontinuity set. Arch. Ration. Mech. Anal. 108(1989), 195–218.

[19] I. Fonseca, G. Leoni. Modern methods in the calculus of variations: Lp

spaces. Springer, New York 2007.

[20] G. A. Francfort, C. J. Larsen. Existence and convergence for quasi-static evolution in brittle fracture. Comm. Pure Appl. Math. 56 (2003),1465–1500.

[21] M. Friedrich. A derivation of linearized Griffith energies from nonlinearmodels. preprint ArXiv: 1504.01671, (2015).

54

[22] M. Friedrich. A Korn-type inequality in SBD for functions with smalljump sets. preprint ArXiv: 1505.00565, (2015).

[23] M. Friedrich. A piecewise Korn inequality in SBD and applications toembedding and density results. preprint ArXiv: 1604.08416, (2016).

[24] M. Friedrich. On a decomposition of regular domains into John domainswith uniform constants. preprint ArXiv: 1605.00130, (2016).

[25] G. A. Francfort, J.-J. Marigo. Revisiting brittle fracture as an energyminimization problem. J. Mech. Phys. Solids 46 (1998), 1319–1342.

[26] A. Giacomini, M. Ponsiglione. Discontinuous finite element approxima-tion of quasi-static growth of brittle fractures. Numer. Funct. Anal. Optim.24 (2003), 813–850.

[27] S. Go lab. Sur quelques points de la theorie de la longueur. Annales de laSociete Polonaise de Mathematiques 7 (1928), 227–241.

[28] A. Griffith. The phenomena of rupture and flow in solids. Phil. Trans.Roy. Soc. London 221-A (1920), 163–198.

[29] F. Iurlano. A density result for GSBD and its application to the approxi-mation of brittle fracture energies. Calc. Var. PDE 51 (2014), 315–342.

[30] F. Maggi. Sets of Finite Perimeter and Geometric Variational Problems:An Introduction to Geometric Measure Theory. Cambridge Studies in Ad-vanced Mathematics No. 135. Cambridge University Press, Cambridge 2012.

[31] D. Mumford, J. Shah. Optimal approximation by piecewise smooth func-tions and associated variational problems. Commun. Pure Appl. Math. 42(1989), 577–684.

[32] F. Murat. The Neumann sieve. Nonlinear variational problems (Isolad’Elba, 1983). Research Notes in Mathematics, 127. Pitman, Boston 1985.

[33] J. A. Nitsche. On Korn’s second inequality. RAIRO Anal. Numer 15(1981), 237–248.

[34] R. Temam. Mathematical problems in plasticity. Gauthier-Villars, Paris1985.

55